Upload folder using huggingface_hub

config.json

@@ -0,0 +1,66 @@
+{
+  "_name_or_path": "/common/home/users/w/wzhao/qllava/vqllava3_finetune_on_filted_dataset_new/checkpoint-2131",
+  "architectures": [
+    "LlavaNextForConditionalGeneration"
+  ],
+  "ignore_index": -100,
+  "image_grid_pinpoints": [
+    [
+      336,
+      672
+    ],
+    [
+      672,
+      336
+    ],
+    [
+      672,
+      672
+    ],
+    [
+      1008,
+      336
+    ],
+    [
+      336,
+      1008
+    ]
+  ],
+  "image_seq_length": 576,
+  "image_token_index": 128256,
+  "model_type": "llava_next",
+  "projector_hidden_act": "gelu",
+  "text_config": {
+    "_name_or_path": "meta-llama/Meta-Llama-3-8B-Instruct",
+    "architectures": [
+      "LlamaForCausalLM"
+    ],
+    "bos_token_id": 128000,
+    "eos_token_id": 128009,
+    "intermediate_size": 14336,
+    "max_position_embeddings": 8192,
+    "model_type": "llama",
+    "num_key_value_heads": 8,
+    "rms_norm_eps": 1e-05,
+    "rope_theta": 500000.0,
+    "torch_dtype": "bfloat16",
+    "vocab_size": 128320
+  },
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.46.1",
+  "use_image_newline_parameter": true,
+  "vision_config": {
+    "hidden_size": 1024,
+    "image_size": 336,
+    "intermediate_size": 4096,
+    "model_type": "clip_vision_model",
+    "num_attention_heads": 16,
+    "num_hidden_layers": 24,
+    "patch_size": 14,
+    "projection_dim": 768,
+    "vocab_size": 32000
+  },
+  "vision_feature_layer": -2,
+  "vision_feature_select_strategy": "default"
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 128000,
+  "eos_token_id": 128009,
+  "transformers_version": "4.46.1"
+}

model-00001-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:316a0b931dd6da3341cbd4f6beb6bbd07b347278d61f03e5bdda86fa050b7e9d
+size 4939760848

model-00002-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fbda1cd72808cce11e02dfe6950690561589bc828b35e0853d26978301382e2
+size 4999813864

model-00003-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84a0f8a111d4d37c215122a185b6cfda626456b618a9ee94c3855fee778f1ff0
+size 4832008176

model-00004-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:486029793e97d9610fc0c2819b8b1785eff787ec1bee69df4380f8f911dcfa62
+size 4999813920

model-00005-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c133428f118b4a785d0c2763544b5694735c3ea90f660ff08c3bbcdf9569627
+size 4999813920

model-00006-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59027fb4ae13d7c242b5cfad81feebdf645499338dc6a7885a548f7ef2228d54
+size 4832008200

model-00007-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b0ea2e7d880e4103b58fb3bb8f2497f1c3118cba658dc32dc1987aaa5d61a17
+size 4082224048

model.safetensors.index.json

@@ -0,0 +1,698 @@
+{
+  "metadata": {
+    "total_size": 33685350400
+  },
+  "weight_map": {
+    "image_newline": "model-00001-of-00007.safetensors",
+    "language_model.lm_head.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.embed_tokens.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.input_layernorm.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.mlp.down_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.mlp.gate_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.mlp.up_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.self_attn.o_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.0.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.1.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.1.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.1.mlp.gate_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.1.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.1.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.1.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.1.self_attn.o_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.1.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.1.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "language_model.model.layers.10.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.10.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.11.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.12.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.12.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.12.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.12.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.12.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.12.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.12.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.12.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.12.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.13.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.13.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.14.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.15.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.16.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.17.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.18.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.18.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.18.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.18.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.18.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.18.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.18.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.18.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.18.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "language_model.model.layers.19.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.19.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.2.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.2.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.20.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.20.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.21.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.22.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.23.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.24.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.24.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.24.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.24.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.24.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.24.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.24.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.24.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.24.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "language_model.model.layers.25.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.25.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.26.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.27.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.28.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.29.input_layernorm.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.29.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.29.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.29.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.29.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.29.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.29.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.29.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.29.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "language_model.model.layers.3.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.3.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.30.input_layernorm.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.mlp.gate_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.mlp.up_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.self_attn.k_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.self_attn.o_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.self_attn.q_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.30.self_attn.v_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.input_layernorm.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.mlp.gate_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.mlp.up_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.self_attn.k_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.self_attn.o_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.self_attn.q_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.31.self_attn.v_proj.weight": "model-00007-of-00007.safetensors",
+    "language_model.model.layers.4.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.4.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.5.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.6.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.7.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.7.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.7.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.7.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.7.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.7.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.7.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.7.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.7.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "language_model.model.layers.8.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.8.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.layers.9.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "language_model.model.norm.weight": "model-00007-of-00007.safetensors",
+    "multi_modal_projector.linear_1.bias": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.linear_1.weight": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.linear_2.bias": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.linear_2.weight": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.vq.embedding.weight": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.vq_cls._category_mapping_indices": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.vq_cls._category_mapping_names": "model-00001-of-00007.safetensors",
+    "multi_modal_projector.vq_cls.embedding.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.embeddings.class_embedding": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.embeddings.patch_embedding.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.embeddings.position_embedding.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.0.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.1.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.10.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.11.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.12.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.13.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.14.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.15.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.16.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.17.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.18.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.19.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.2.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.20.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.21.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.22.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.23.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.3.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.4.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.5.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.6.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.7.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.8.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.layer_norm1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.layer_norm1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.layer_norm2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.layer_norm2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.mlp.fc1.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.mlp.fc1.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.mlp.fc2.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.mlp.fc2.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.k_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.out_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.out_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.q_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.v_proj.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.encoder.layers.9.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.post_layernorm.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.post_layernorm.weight": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.pre_layrnorm.bias": "model-00001-of-00007.safetensors",
+    "vision_tower.vision_model.pre_layrnorm.weight": "model-00001-of-00007.safetensors"
+  }
+}

qllava3_test.py

@@ -0,0 +1,1466 @@
+# coding=utf-8
+# Copyright 2024 the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""PyTorch Llava-NeXT model."""
+
+import math
+from dataclasses import dataclass
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+import torch.utils.checkpoint
+from torch import nn
+import torch.nn.functional as F
+
+from transformers.activations import ACT2FN
+from transformers.generation import GenerationMixin
+from transformers.image_processing_utils import select_best_resolution
+from transformers.modeling_outputs import ModelOutput
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+from transformers.models.auto import AutoModel, AutoModelForCausalLM
+from transformers.models.llava_next.configuration_llava_next import LlavaNextConfig
+
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "LlavaNextConfig"
+from pathlib import Path
+
+def save_list_to_incremental_file(data_list, save_dir="/common/home/users/w/wzhao/vqclip/llava_next_tensors"):
+    """
+    将列表保存到指定目录，文件名按数字递增
+    
+    Args:
+        data_list: 要保存的列表数据
+        save_dir: 保存目录路径
+    
+    Returns:
+        保存的文件路径
+    """
+    # 确保目录存在
+    save_dir = Path(save_dir)
+    save_dir.mkdir(parents=True, exist_ok=True)
+    
+    # 查找可用的文件名
+    index = 1
+    while True:
+        file_path = save_dir / f"{index}.npy"
+        if not file_path.exists():
+            break
+        index += 1
+    
+    # 将列表转换为numpy数组并保存
+    np_array = np.array(data_list)
+    np.save(str(file_path), np_array)
+    
+    return file_path
+
+def get_anyres_image_grid_shape(image_size, grid_pinpoints, patch_size):
+    """
+    Calculate the shape of the image patch grid after the preprocessing for images of any resolution.
+
+    Args:
+        image_size (`tuple`):
+            The size of the input image in the format (width, height).
+        grid_pinpoints (`List`):
+            A list containing possible resolutions. Each item in the list should be a tuple or list
+            of the form `(height, width)`.
+        patch_size (`int`):
+            The size of each image patch.
+
+    Returns:
+        tuple: The shape of the image patch grid in the format (width, height).
+    """
+    if not isinstance(grid_pinpoints, list):
+        raise TypeError("grid_pinpoints should be a list of tuples or lists")
+
+    # ! VERY IMPORTANT if image_size is tensor, must convert to into tuple, otherwise it will cause wrong calculate
+    if not isinstance(image_size, (list, tuple)):
+        if not isinstance(image_size, (torch.Tensor, np.ndarray)):
+            raise TypeError(
+                f"image_size invalid type: {type(image_size)} not valid, should be either list, tuple, np.ndarray or tensor"
+            )
+        image_size = image_size.tolist()
+
+    height, width = select_best_resolution(image_size, grid_pinpoints)
+    return height // patch_size, width // patch_size
+
+
+def image_size_to_num_patches(image_size, grid_pinpoints, patch_size: int):
+    """
+    Calculate the number of patches after the preprocessing for images of any resolution.
+
+    Args:
+        image_size (`torch.LongTensor` or `np.ndarray` or `Tuple[int, int]`):
+            The size of the input image in the format (height, width). ?
+        grid_pinpoints (`List`):
+            A list containing possible resolutions. Each item in the list should be a tuple or list
+            of the form `(height, width)`.
+        patch_size (`int`):
+            The size of each image patch.
+
+    Returns:
+        int: the number of patches
+    """
+    if not isinstance(grid_pinpoints, list):
+        raise TypeError("grid_pinpoints should be a list of tuples or lists")
+
+    # ! VERY IMPORTANT if image_size is tensor, must convert to into tuple, otherwise it will cause wrong calculate
+    if not isinstance(image_size, (list, tuple)):
+        if not isinstance(image_size, (torch.Tensor, np.ndarray)):
+            raise TypeError(f"image_size invalid type {type(image_size)} with value {image_size}")
+        image_size = image_size.tolist()
+
+    best_resolution = select_best_resolution(image_size, grid_pinpoints)
+    height, width = best_resolution
+    num_patches = 0
+    # consider change to ceil(height/patch_size)*ceil(width/patch_size) + 1
+    for i in range(0, height, patch_size):
+        for j in range(0, width, patch_size):
+            num_patches += 1
+    # add the base patch
+    num_patches += 1
+    return num_patches
+
+
+def unpad_image(tensor, original_size):
+    """
+    Unpads a PyTorch tensor of a padded and resized image.
+
+    Args:
+        tensor (`torch.Tensor`):
+            The image tensor, assumed to be of shape (num_channels, height, width).
+        original_size (`tuple`):
+            The original size of the image (height, width).
+
+    Returns:
+        `torch.Tensor`: The unpadded image tensor.
+    """
+    if not isinstance(original_size, (list, tuple)):
+        if not isinstance(original_size, (torch.Tensor, np.ndarray)):
+            raise TypeError(
+                f"image_size invalid type: {type(original_size)} not valid, should be either list, tuple, np.ndarray or tensor"
+            )
+        original_size = original_size.tolist()
+    original_height, original_width = original_size
+    current_height, current_width = tensor.shape[1:]
+
+    original_aspect_ratio = original_width / original_height
+    current_aspect_ratio = current_width / current_height
+
+    if original_aspect_ratio > current_aspect_ratio:
+        scale_factor = current_width / original_width
+        new_height = int(original_height * scale_factor)
+        padding = (current_height - new_height) // 2
+        unpadded_tensor = tensor[:, padding : current_height - padding, :]
+    else:
+        scale_factor = current_height / original_height
+        new_width = int(original_width * scale_factor)
+        padding = (current_width - new_width) // 2
+        unpadded_tensor = tensor[:, :, padding : current_width - padding]
+
+    return unpadded_tensor
+
+
+@dataclass
+class LlavaNextCausalLMOutputWithPast(ModelOutput):
+    """
+    Base class for LlavaNext causal language model (or autoregressive) outputs.
+
+    Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Language modeling loss (for next-token prediction).
+        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
+            `past_key_values` input) to speed up sequential decoding.
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+        image_hidden_states (`torch.FloatTensor`, *optional*):
+            A `torch.FloatTensor` of size (batch_size * num_patches, num_images, sequence_length, hidden_size)`.
+            image_hidden_states of the model produced by the vision encoder and after projecting the last hidden state.
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    logits: torch.FloatTensor = None
+    past_key_values: Optional[List[torch.FloatTensor]] = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+    image_hidden_states: Optional[torch.FloatTensor] = None
+
+class VectorQuantizer(nn.Module):
+    def __init__(self, num_embeddings: int, embedding_dim: int, commitment_cost: float = 0.25):
+        super().__init__()
+        self.num_embeddings = num_embeddings
+        self.embedding_dim = embedding_dim
+        self.commitment_cost = commitment_cost
+        
+        # Embedding table
+        self.embedding = nn.Embedding(num_embeddings, embedding_dim)
+        self.embedding.weight.data.uniform_(-1/num_embeddings, 1/num_embeddings)
+
+    def forward(self, inputs):
+        
+        self.embedding.weight.data = self.embedding.weight.data.to(dtype=inputs.dtype)
+        # Convert inputs from BCHW -> BHWC
+        inputs = inputs.permute(0, 2, 1).contiguous()
+        input_shape = inputs.shape
+        
+        # Flatten input
+        flat_input = inputs.view(-1, self.embedding_dim)
+
+        # Calculate distances
+        distances = (torch.sum(flat_input**2, dim=1, keepdim=True) 
+                    + torch.sum(self.embedding.weight**2, dim=1)
+                    - 2 * torch.matmul(flat_input, self.embedding.weight.t()))
+            
+        # Encoding
+        encoding_indices = torch.argmin(distances, dim=1).unsqueeze(1)
+        encodings = torch.zeros(encoding_indices.shape[0], self.num_embeddings, device=inputs.device).to(inputs.dtype)
+        encodings.scatter_(1, encoding_indices, 1)
+        #self.embedding.weight = self.embedding.weight.to(input_type)
+        # Quantize and unflatten
+        #print(inputs.dtype)
+        #print(self.embedding.weight.dtype)
+        quantized = torch.matmul(encodings, self.embedding.weight).view(input_shape)
+        
+        # Loss
+        e_latent_loss = torch.mean((quantized.detach() - inputs)**2)
+        q_latent_loss = torch.mean((quantized - inputs.detach())**2)
+        loss = q_latent_loss + self.commitment_cost * e_latent_loss
+        print("this is q_latent_loss", q_latent_loss)
+        print("This is e_latent_loss", self.commitment_cost * e_latent_loss)
+        quantized = inputs + (quantized - inputs).detach()
+        avg_probs = torch.mean(encodings, dim=0)
+        perplexity = torch.exp(-torch.sum(avg_probs * torch.log(avg_probs + 1e-10)))
+
+        # Convert quantized from BHWC -> BCHW
+        return quantized.permute(0, 2, 1).contiguous(), loss, perplexity
+
+# Copied from transformers.models.llava.modeling_llava.LlavaMultiModalProjector with Llava->LlavaNext
+class LlavaNextMultiModalProjector(nn.Module):
+    def __init__(self, config: LlavaNextConfig):
+        super().__init__()
+
+        self.linear_1 = nn.Linear(config.vision_config.hidden_size, config.text_config.hidden_size, bias=True)
+        self.act = ACT2FN[config.projector_hidden_act]
+        self.linear_2 = nn.Linear(config.text_config.hidden_size, config.text_config.hidden_size, bias=True)
+        self.vq = VectorQuantizer(
+                    num_embeddings=16000,  # codebook size
+                    embedding_dim=config.text_config.hidden_size,  # dimension of each embedding vector
+                    commitment_cost=0.5
+                )
+        self.vq_cls = VectorQuantizerCLS(
+            num_embeddings=128,
+            embedding_dim=4096,
+            commitment_cost=0.25,
+            use_cosine=True
+        )
+    def forward(self, image_features):
+        cls_features = image_features[: , :1]
+        cls_features = self.linear_1(cls_features)
+        cls_features = self.act(cls_features)
+        cls_features = self.linear_2(cls_features)
+        cls_features = cls_features[:, 0:]
+        cls_features = cls_features.mean(dim=0, keepdim=True).squeeze(0)
+        #save_list_to_incremental_file(cls_features.cpu().detach().numpy())
+        quantized, loss, perplexity, indices = self.vq_cls(cls_features)
+        categories = self.vq_cls.get_category_from_index(indices)
+        indices = indices.cpu().numpy()
+        print(indices)
+        print(categories)
+        if categories[0] != 0:
+            raise ValueError([indices, categories[0]]) 
+        #save_list_to_incremental_file(save_list)
+        # tensor(54)
+        # ['porn']
+        image_features = image_features[: , 1:]
+        hidden_states = self.linear_1(image_features)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.linear_2(hidden_states)
+        
+        quantized_features, vq_loss, perplexity = self.vq(hidden_states)
+        print(quantized_features.shape)
+        return quantized_features, vq_loss
+
+
+LLAVA_NEXT_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`LlavaNextConfig`] or [`LlavaNextVisionConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+class VectorQuantizerCLS(nn.Module):
+    def __init__(self, num_embeddings: int = 64, embedding_dim: int = 4096, commitment_cost: float = 0.25, 
+                 codebook_path: str = None, mapping_path: str = None, use_cosine: bool = True, 
+                 randomize_indices: bool = True):
+        super().__init__()
+        self.num_embeddings = num_embeddings
+        self.embedding_dim = embedding_dim
+        self.commitment_cost = commitment_cost
+        self.use_cosine = use_cosine
+        
+        # Embedding table
+        self.embedding = nn.Embedding(num_embeddings, embedding_dim)
+        self.embedding.weight.data.uniform_(-1/num_embeddings, 1/num_embeddings)
+        
+        # 初始化合适大小的buffer，避免加载时大小不匹配
+        self.register_buffer('_category_mapping_indices', torch.zeros(num_embeddings, dtype=torch.long))
+        self.register_buffer('_category_mapping_names', torch.zeros(num_embeddings, dtype=torch.long))
+        
+        # 非持久化属性
+        self.center_to_category = None
+        
+        # 加载预先计算的codebook
+        if codebook_path is not None and mapping_path is not None:
+            self.load_codebook(codebook_path, mapping_path, randomize_indices)
+    
+    def load_codebook(self, codebook_path, mapping_path, randomize_indices=True):
+        """加载预计算的codebook和类别映射，并可选择随机化索引"""
+        try:
+            # 加载codebook
+            print(f"Loading codebook from {codebook_path}")
+            centers = np.load(codebook_path)
+            print(f"Loaded codebook with shape: {centers.shape}")
+            
+            # 加载类别映射
+            print(f"Loading category mappings from {mapping_path}")
+            with open(mapping_path, 'rb') as f:
+                mappings = pickle.load(f)
+            
+            # 将文本类别映射为数字
+            category_mapping_text = mappings['category_mapping']
+            classes = {'neutral':0, 'porn':1, 'gun':2, 'cigarette':3, 'alcohol':4, 'knife':5, 'blood':6, 'insulting_gesture':7}
+            
+            # 转换为数字映射
+            center_category_mapping = {}
+            for i, category_text in enumerate(category_mapping_text):
+                center_category_mapping[i] = classes.get(category_text, 0)  # 默认为neutral(0)
+            
+            print(f"Loaded {len(center_category_mapping)} category mappings")
+            
+            # 准备数据
+            actual_centers = centers.shape[0]
+            print(f"Actual centers: {actual_centers}")
+            
+            # 更新num_embeddings为实际中心点数量
+            self.num_embeddings = actual_centers
+            print(f"Setting num_embeddings to {self.num_embeddings}")
+            
+            # 如果需要随机化索引，创建随机排列
+            if randomize_indices:
+                print("Randomizing codebook indices to prevent category clustering")
+                # 创建随机排列
+                permutation = list(range(actual_centers))
+                random.shuffle(permutation)
+                inverse_permutation = {v: k for k, v in enumerate(permutation)}
+                
+                # 应用随机排列到中心点和类别映射
+                permuted_centers = np.zeros_like(centers)
+                permuted_categories = {}
+                
+                for new_idx, old_idx in enumerate(permutation):
+                    permuted_centers[new_idx] = centers[old_idx]
+                    if old_idx < len(center_category_mapping):
+                        permuted_categories[new_idx] = center_category_mapping[old_idx]
+                
+                # 使用随机化后的数据
+                centers = permuted_centers
+                self.center_to_category = permuted_categories
+                
+                # 打印一些随机化后的映射示例
+                print("Sample randomized mappings:")
+                for i in range(min(5, len(self.center_to_category))):
+                    print(f"  New index {i}: {self.center_to_category[i]}")
+            else:
+                # 不随机化，直接使用原始映射
+                self.center_to_category = {i: center_category_mapping[i] 
+                                          for i in range(min(actual_centers, len(center_category_mapping)))}
+            
+            # 验证类别映射是否完整
+            for i in range(self.num_embeddings):
+                if i not in self.center_to_category:
+                    print(f"Warning: No category mapping for center {i}, setting to 0")
+                    self.center_to_category[i] = 0  # 用0代替"unknown"
+            
+            # 创建embedding数据并更新
+            embedding_data = torch.tensor(centers, dtype=torch.float32)
+            
+            # 重新创建embedding层以匹配实际大小
+            self.embedding = nn.Embedding(self.num_embeddings, self.embedding_dim)
+            self.embedding.weight.data.copy_(embedding_data)
+            
+            # 重新注册buffer以匹配新的大小
+            self.register_buffer('_category_mapping_indices', torch.zeros(self.num_embeddings, dtype=torch.long))
+            self.register_buffer('_category_mapping_names', torch.zeros(self.num_embeddings, dtype=torch.long))
+            
+            # 将类别映射存储到buffer中（用于state_dict）
+            self._store_category_mapping()
+            
+            print(f"Successfully loaded codebook with {self.num_embeddings} entries")
+            
+            # 分析类别分布
+            category_counts = {}
+            for category in self.center_to_category.values():
+                if category in category_counts:
+                    category_counts[category] += 1
+                else:
+                    category_counts[category] = 1
+            
+            print("Category distribution in codebook:")
+            for category, count in sorted(category_counts.items()):
+                print(f"  {category}: {count} centers")
+            
+            return True
+            
+        except Exception as e:
+            print(f"Error loading codebook: {e}")
+            import traceback
+            traceback.print_exc()
+            print("Using random initialization instead")
+            return False
+    
+    def _store_category_mapping(self):
+        """将类别映射存储到模型的buffer中，以便在state_dict中保存"""
+        if not self.center_to_category:
+            warnings.warn("No category mapping to store")
+            return
+        
+        # 获取所有类别ID
+        all_categories = sorted(set(self.center_to_category.values()))
+        
+        # 创建索引和对应类别ID的映射
+        indices = list(self.center_to_category.keys())
+        category_ids = [self.center_to_category[idx] for idx in indices]
+        
+        # 确保indices数组长度与buffer大小一致
+        if len(indices) != self._category_mapping_indices.size(0):
+            # 重新注册buffer以匹配大小
+            self.register_buffer('_category_mapping_indices', torch.zeros(len(indices), dtype=torch.long))
+            self.register_buffer('_category_mapping_names', torch.zeros(len(indices), dtype=torch.long))
+        
+        # 存储到buffer中
+        self._category_mapping_indices.copy_(torch.tensor(indices, dtype=torch.long))
+        self._category_mapping_names.copy_(torch.tensor(category_ids, dtype=torch.long))
+        
+        print(f"Stored category mapping with {len(indices)} entries and {len(all_categories)} unique categories")
+    
+    def _load_category_mapping(self):
+        """从模型的buffer恢复类别映射"""
+        if not hasattr(self, '_category_mapping_indices') or self._category_mapping_indices.numel() == 0:
+            warnings.warn("No stored category mapping found")
+            return {}
+        
+        # 重建类别映射字典
+        indices = self._category_mapping_indices.tolist()
+        category_ids = self._category_mapping_names.tolist()
+        
+        mapping = {}
+        for idx, cat_id in zip(indices, category_ids):
+            mapping[idx] = cat_id
+        
+        return mapping
+    
+    def _load_from_state_dict(self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs):
+        """自定义state_dict加载方法，处理buffer大小不匹配的问题"""
+        # 检查并调整buffer大小，以匹配加载的state_dict
+        indices_key = prefix + '_category_mapping_indices'
+        names_key = prefix + '_category_mapping_names'
+        
+        if indices_key in state_dict and names_key in state_dict:
+            indices_size = state_dict[indices_key].size()
+            names_size = state_dict[names_key].size()
+            
+            # 重新注册buffer以匹配加载的大小
+            if hasattr(self, '_category_mapping_indices') and self._category_mapping_indices.size() != indices_size:
+                self.register_buffer('_category_mapping_indices', torch.zeros(indices_size, dtype=torch.long))
+            
+            if hasattr(self, '_category_mapping_names') and self._category_mapping_names.size() != names_size:
+                self.register_buffer('_category_mapping_names', torch.zeros(names_size, dtype=torch.long))
+        
+        # 调用父类方法加载常规参数
+        super()._load_from_state_dict(state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs)
+        
+        # 在加载完成后重建类别映射
+        self.center_to_category = self._load_category_mapping()
+        
+        # 更新num_embeddings以匹配加载的模型
+        if hasattr(self, 'embedding') and hasattr(self.embedding, 'weight'):
+            self.num_embeddings = self.embedding.weight.size(0)
+    
+    def forward(self, inputs):
+        """
+        前向传播，专门处理(1, 4096)形状的输入
+        
+        Args:
+            inputs: 形状为(1, 4096)的特征向量
+            
+        Returns:
+            quantized: 量化后的特征向量
+            loss: 承诺损失
+            perplexity: 困惑度
+            encoding_indices: 编码索引
+        """
+        # 验证输入形状
+        if inputs.shape != (1, 4096):
+            raise ValueError(f"Expected input shape (1, 4096), got {inputs.shape}")
+        
+        # 确保embedding权重与输入使用相同的数据类型
+        self.embedding.weight.data = self.embedding.weight.data.to(dtype=inputs.dtype)
+        
+        # 直接使用输入，不需要形状转换
+        flat_input = inputs
+        
+        # 计算与codebook中各向量的距离
+        if self.use_cosine:
+            # 归一化向量进行余弦相似度计算
+            normalized_input = F.normalize(flat_input, p=2, dim=1)
+            normalized_weights = F.normalize(self.embedding.weight, p=2, dim=1)
+            
+            # 计算余弦相似度
+            cosine_sim = torch.matmul(normalized_input, normalized_weights.t())
+            
+            # 将相似度转换为距离（最大相似度对应最小距离）
+            distances = 1 - cosine_sim
+        else:
+            # 使用欧氏距离
+            distances = (torch.sum(flat_input**2, dim=1, keepdim=True) 
+                        + torch.sum(self.embedding.weight**2, dim=1)
+                        - 2 * torch.matmul(flat_input, self.embedding.weight.t()))
+        
+        # 找到最近的编码向量索引
+        encoding_indices = torch.argmin(distances, dim=1).unsqueeze(1)
+        
+        # 创建one-hot编码
+        encodings = torch.zeros(encoding_indices.shape[0], self.num_embeddings, device=inputs.device).to(inputs.dtype)
+        encodings.scatter_(1, encoding_indices, 1)
+        
+        # 量化
+        quantized = torch.matmul(encodings, self.embedding.weight)
+        
+        # 计算损失
+        e_latent_loss = torch.mean((quantized.detach() - flat_input)**2)
+        q_latent_loss = torch.mean((quantized - flat_input.detach())**2)
+        loss = q_latent_loss + self.commitment_cost * e_latent_loss
+        
+        print("this is q_latent_loss", q_latent_loss)
+        print("This is e_latent_loss", self.commitment_cost * e_latent_loss)
+        
+        # Straight-through estimator
+        quantized = flat_input + (quantized - flat_input).detach()
+        
+        # 计算perplexity
+        avg_probs = torch.mean(encodings, dim=0)
+        perplexity = torch.exp(-torch.sum(avg_probs * torch.log(avg_probs + 1e-10)))
+        
+        # 返回量化后的向量、损失、困惑度和索引
+        return quantized, loss, perplexity, encoding_indices.squeeze()
+    
+    def encode(self, inputs):
+        """
+        仅执行编码过程，返回索引
+        
+        Args:
+            inputs: 形状为(1, 4096)的特征向量
+            
+        Returns:
+            编码索引
+        """
+        # 验证输入形状
+        if inputs.shape != (1, 4096):
+            raise ValueError(f"Expected input shape (1, 4096), got {inputs.shape}")
+        
+        with torch.no_grad():
+            # 计算距离
+            if self.use_cosine:
+                normalized_input = F.normalize(inputs, p=2, dim=1)
+                normalized_weights = F.normalize(self.embedding.weight, p=2, dim=1)
+                cosine_sim = torch.matmul(normalized_input, normalized_weights.t())
+                distances = 1 - cosine_sim
+            else:
+                distances = (torch.sum(inputs**2, dim=1, keepdim=True) 
+                            + torch.sum(self.embedding.weight**2, dim=1)
+                            - 2 * torch.matmul(inputs, self.embedding.weight.t()))
+            
+            # 找到最近的编码向量索引
+            encoding_indices = torch.argmin(distances, dim=1)
+            
+            return encoding_indices
+    
+    def get_category_from_index(self, indices):
+        """
+        根据索引获取对应的类别编号
+        
+        Args:
+            indices: 编码索引
+            
+        Returns:
+            类别编号列表
+        """
+        # 如果没有类别映射，尝试从buffer恢复
+        if self.center_to_category is None:
+            self.center_to_category = self._load_category_mapping()
+            
+        if not self.center_to_category:
+            return [0] * indices.numel()  # 使用0(neutral)代替"unknown"
+        
+        # 将索引张量转为NumPy数组
+        indices_np = indices.cpu().numpy().flatten()
+        
+        # 获取类别
+        categories = []
+        for idx in indices_np:
+            idx_int = int(idx)
+            category = self.center_to_category.get(idx_int, 0)  # 默认为0(neutral)
+            categories.append(category)
+        
+        return categories
+    
+    def classify(self, inputs):
+        """
+        对输入特征进行分类，返回类别编号和索引
+        
+        Args:
+            inputs: 形状为(1, 4096)的特征向量
+            
+        Returns:
+            categories: 预测的类别编号
+            indices: 编码索引
+        """
+        # 验证输入形状
+        if inputs.shape != (1, 4096):
+            raise ValueError(f"Expected input shape (1, 4096), got {inputs.shape}")
+        
+        indices = self.encode(inputs)
+        categories = self.get_category_from_index(indices)
+        return categories, indices
+
+
+@add_start_docstrings(
+    "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+    LLAVA_NEXT_START_DOCSTRING,
+)
+# Copied from transformers.models.llava.modeling_llava.LlavaPreTrainedModel with Llava->LlavaNext,llava->llava_next
+class LlavaNextPreTrainedModel(PreTrainedModel):
+    config_class = LlavaNextConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["LlavaNextVisionAttention"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_cache_class = True
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+
+    def _init_weights(self, module):
+        # important: this ported version of LlavaNext isn't meant for training from scratch - only
+        # inference and fine-tuning - so the proper init weights code has been removed - the original codebase
+        # https://github.com/haotian-liu/LLaVA/tree/main/llava_next should serve for that purpose
+        std = (
+            self.config.initializer_range
+            if hasattr(self.config, "initializer_range")
+            else self.config.text_config.initializer_range
+        )
+
+        if hasattr(module, "class_embedding"):
+            module.class_embedding.data.normal_(mean=0.0, std=std)
+
+        if isinstance(module, (nn.Linear, nn.Conv2d)):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+
+LLAVA_NEXT_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, image_size, image_size)):
+            The tensors corresponding to the input images. Pixel values can be obtained using
+            [`AutoImageProcessor`]. See [`LlavaNextImageProcessor.__call__`] for details. [`LlavaProcessor`] uses
+            [`LlavaNextImageProcessor`] for processing images.
+        image_sizes (`torch.LongTensor` of shape `(batch_size, 2)`, *optional*):
+            The sizes of the images in the batch, being (height, width) for each image.
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`. [What are position IDs?](../glossary#position-ids)
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        vision_feature_layer (`int`, *optional*, defaults to -2):
+            The index of the layer to select the vision feature.
+        vision_feature_select_strategy (`str`, *optional*, defaults to `"default"`):
+            The feature selection strategy used to select the vision feature from the vision backbone.
+            Can be one of `"default"` or `"full"`. If `"default"`, the CLS token is removed from the vision features.
+            If `"full"`, the full vision features are used.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
+            Indices depicting the position of the input sequence tokens in the sequence. Contrarily to `position_ids`,
+            this tensor is not affected by padding. It is used to update the cache in the correct position and to infer
+            the complete sequence length.
+"""
+
+
+@add_start_docstrings(
+    """The LLAVA-NeXT model which consists of a vision backbone and a language model.""",
+    LLAVA_NEXT_START_DOCSTRING,
+)
+class LlavaNextForConditionalGeneration(LlavaNextPreTrainedModel, GenerationMixin):
+    def __init__(self, config: LlavaNextConfig):
+        super().__init__(config)
+        self.vision_tower = AutoModel.from_config(config.vision_config)
+
+        self.multi_modal_projector = LlavaNextMultiModalProjector(config)
+        embed_std = 1 / math.sqrt(config.text_config.hidden_size)
+        self.image_newline = nn.Parameter(torch.randn(config.text_config.hidden_size, dtype=self.dtype) * embed_std)
+
+        self.vocab_size = config.text_config.vocab_size
+        self.language_model = AutoModelForCausalLM.from_config(config.text_config)
+        self.pad_token_id = self.config.pad_token_id if self.config.pad_token_id is not None else -1
+        self._padding_side = "left"  # set it to left by default, user can use setter to change padding_sides
+        self.post_init()
+
+    @property
+    def padding_side(self):
+        return self._padding_side
+
+    @padding_side.setter
+    def padding_side(self, padding_side: str):
+        if padding_side not in ["left", "right"]:
+            raise ValueError(f"{padding_side} is not `left` or `right`.")
+        self._padding_side = padding_side
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.get_input_embeddings
+    def get_input_embeddings(self):
+        return self.language_model.get_input_embeddings()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.set_input_embeddings
+    def set_input_embeddings(self, value):
+        self.language_model.set_input_embeddings(value)
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.get_output_embeddings
+    def get_output_embeddings(self):
+        return self.language_model.get_output_embeddings()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.set_output_embeddings
+    def set_output_embeddings(self, new_embeddings):
+        self.language_model.set_output_embeddings(new_embeddings)
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.set_decoder
+    def set_decoder(self, decoder):
+        self.language_model.set_decoder(decoder)
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.get_decoder
+    def get_decoder(self):
+        return self.language_model.get_decoder()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.tie_weights
+    def tie_weights(self):
+        return self.language_model.tie_weights()
+
+    # Copied from transformers.models.llava.modeling_llava.LlavaForConditionalGeneration.resize_token_embeddings
+    def resize_token_embeddings(self, new_num_tokens: Optional[int] = None, pad_to_multiple_of=None) -> nn.Embedding:
+        model_embeds = self.language_model.resize_token_embeddings(new_num_tokens, pad_to_multiple_of)
+        # update vocab size
+        self.config.text_config.vocab_size = model_embeds.num_embeddings
+        self.vocab_size = model_embeds.num_embeddings
+        return model_embeds
+
+    def _merge_input_ids_with_image_features(
+        self,
+        image_features,
+        feature_lens,
+        inputs_embeds,
+        input_ids,
+        attention_mask,
+        position_ids=None,
+        labels=None,
+        image_token_index=None,
+        ignore_index=-100,
+    ):
+        """
+        Merge input_ids with with image features into final embeddings
+
+        Args:
+            image_features (`torch.Tensor` of shape `(all_feature_lens, embed_dim)`):
+                All vision vectors of all images in the batch
+            feature_lens (`torch.LongTensor` of shape `(num_images)`):
+                The length of visual embeddings of each image as stacked in `image_features`
+            inputs_embeds (`torch.Tensor` of shape `(batch_size, sequence_length, embed_dim)`):
+                Token embeddings before merging with visual embeddings
+            input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Input_ids of tokens, possibly filled with image token
+            attention_mask (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Mask to avoid performing attention on padding token indices.
+            position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+                Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+                config.n_positions - 1]`.
+            labels (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*)
+                :abels need to be recalculated to support training (if provided)
+            image_token_index (`int`, *optional*)
+                Token id used to indicate the special "image" token. Defaults to `config.image_token_index`
+            ignore_index (`int`, *optional*)
+                Value that is used to pad `labels` and will be ignored when calculated loss. Default: -100.
+        Returns:
+            final_embedding, final_attention_mask, position_ids, final_labels
+
+        Explanation:
+            each image has variable length embeddings, with length specified by feature_lens
+            image_features is concatenation of all visual embed vectors
+            task: fill each <image> with the correct number of visual embeddings
+            Example:
+                X (5 patches), Y (3 patches), Z (8)
+                X, Y are in the same sequence (in-context learning)
+            if right padding
+                input_ids: [
+                    a b c d e f X g h i j k Y l m
+                    o p q r Z s t u v _ _ _ _ _ _
+                ]
+                input_ids should be: [
+                    a b c d e f X X X X X g h i j k Y Y Y l m
+                    o p q r Z Z Z Z Z Z Z Z s t u v _ _ _ _ _
+                ]
+                labels should be: [
+                    a b c d e f _ _ _ _ _ g h i j k _ _ _ l m
+                    o p q r _ _ _ _ _ _ _ _ s t u v _ _ _ _ _
+                ]
+            elif left padding
+                input_ids: [
+                    a b c d e f X g h i j k Y l m
+                    _ _ _ _ _ _ o p q r Z s t u v
+                ]
+                input_ids should be: [
+                    a b c d e f X X X X X g h i j k Y Y Y l m
+                    _ _ _ _ _ o p q r Z Z Z Z Z Z Z Z s t u v
+                ]
+                labels should be: [
+                    a b c d e f _ _ _ _ _ g h i j k _ _ _ l m
+                    _ _ _ _ _ o p q r _ _ _ _ _ _ _ _ s t u v
+                ]
+            Edge cases:
+                * If tokens are same but image token sizes are different, then cannot infer left or right padding
+                ```python
+                cat_img = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
+                chart_img = Image.open(requests.get("https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true", stream=True).raw)
+                prompts = [
+                    "[INST] <image>\nWhat is shown in this image? [/INST]",
+                    "[INST] <image>\nWhat is shown in this image? [/INST]",
+                ]
+                inputs = processor(prompts, [chart_img, cat_img], return_tensors='pt', padding=True).to("cuda")
+                    chart_img has 2634 tokens, while cat_img has 2340 tokens
+                ```
+
+                input_ids: [
+                    a b c d X g h
+                    i j Y k l m n
+                ]
+                where X is 3 tokens while Y is 5, this mean after merge
+                if left-padding (batched generation)
+                    input_ids should be: [
+                        _ _ a b c d X X X g h
+                        i j Y Y Y Y Y k l m n
+                    ]
+                elif (right padding) (training)
+                    input_ids should be: [
+                        a b c d X X X g h _ _
+                        i j Y Y Y Y Y k l m n
+                    ]
+        """
+        image_token_index = image_token_index if image_token_index is not None else self.config.image_token_index
+        ignore_index = ignore_index if ignore_index is not None else self.config.ignore_index
+
+        if self.training and self.padding_side == "left":
+            logger.warning_once(
+                "Padding side is set to 'left' but the model is in training mode. For training "
+                "it is recommended to set `model.padding_side='right' and `processor.tokenizer.padding_side='right'`. "
+                "If that's intended, ignore this warning"
+            )
+        if not self.training and self.padding_side == "right":
+            logger.warning_once(
+                "Padding side is set to 'right' but the model is in inference mode. For correct "
+                "generation results, please set `model.padding_side='left'` and `processor.tokenizer.padding_side='left'`. "
+                "If that's intended, ignore this warning"
+            )
+
+        with torch.no_grad():
+            # ! in llava 1.6, number of patches is variable
+            num_images = feature_lens.size(0)
+            num_image_features, embed_dim = image_features.shape
+            if feature_lens.sum() != num_image_features:
+                raise ValueError(f"{feature_lens=} / {feature_lens.sum()} != {image_features.shape=}")
+            batch_size = input_ids.shape[0]
+            _left_padding = torch.any(attention_mask[:, 0] == 0)
+            _right_padding = torch.any(attention_mask[:, -1] == 0)
+
+            left_padding = self.padding_side == "left"
+            if batch_size > 1:
+                if _left_padding and _right_padding:
+                    raise ValueError(f"both side of attention_mask has zero, invalid. {attention_mask}")
+                elif _right_padding and left_padding:
+                    left_padding = False
+                elif _left_padding and not left_padding:
+                    left_padding = True
+
+            # Whether to turn off right padding
+            # 1. Create a mask to know where special image tokens are
+            special_image_token_mask = input_ids == image_token_index
+            # special_image_token_mask: [bsz, seqlen]
+            num_special_image_tokens = torch.sum(special_image_token_mask, dim=-1)
+            # num_special_image_tokens: [bsz]
+            # Reserve for padding of num_images
+            total_num_special_image_tokens = torch.sum(special_image_token_mask)
+            if total_num_special_image_tokens != num_images:
+                raise ValueError(
+                    f"Number of image tokens in input_ids ({total_num_special_image_tokens}) different from num_images ({num_images})."
+                )
+            # Compute the maximum embed dimension
+            # max_image_feature_lens is max_feature_lens per batch
+            feature_lens = feature_lens.to(input_ids.device)
+            feature_lens_batch = feature_lens.split(num_special_image_tokens.tolist(), dim=0)
+            feature_lens_batch_sum = torch.tensor([x.sum() for x in feature_lens_batch], device=input_ids.device)
+            embed_sequence_lengths = (
+                (attention_mask == 1).long().sum(-1) - num_special_image_tokens + feature_lens_batch_sum
+            )
+            max_embed_dim = embed_sequence_lengths.max()
+
+            batch_indices, non_image_indices = torch.where((input_ids != image_token_index) & (attention_mask == 1))
+            # 2. Compute the positions where text should be written
+            # Calculate new positions for text tokens in merged image-text sequence.
+            # `special_image_token_mask` identifies image tokens. Each image token will be replaced by `nb_text_tokens_per_images` text tokens.
+            # `torch.cumsum` computes how each image token shifts subsequent text token positions.
+            # - 1 to adjust for zero-based indexing, as `cumsum` inherently increases indices by one.
+            # ! instead of special_image_token_mask * (num_image_patches - 1)
+            #   special_image_token_mask * (num_feature_len - 1)
+            special_image_token_mask = special_image_token_mask.long()
+            special_image_token_mask[special_image_token_mask == 1] = feature_lens - 1
+            new_token_positions = torch.cumsum((special_image_token_mask + 1), -1) - 1
+            if left_padding:
+                # shift right token positions so that they are ending at the same number
+                # the below here was incorrect? new_token_positions += new_token_positions[:, -1].max() - new_token_positions[:, -1:]
+                new_token_positions += max_embed_dim - 1 - new_token_positions[:, -1:]
+
+            text_to_overwrite = new_token_positions[batch_indices, non_image_indices]
+
+        # 3. Create the full embedding, already padded to the maximum position
+        final_embedding = torch.zeros(
+            batch_size, max_embed_dim, embed_dim, dtype=inputs_embeds.dtype, device=inputs_embeds.device
+        )
+        final_attention_mask = torch.zeros(
+            batch_size, max_embed_dim, dtype=attention_mask.dtype, device=inputs_embeds.device
+        )
+        final_input_ids = torch.full(
+            (batch_size, max_embed_dim), self.pad_token_id, dtype=input_ids.dtype, device=inputs_embeds.device
+        )
+        # In case the Vision model or the Language model has been offloaded to CPU, we need to manually
+        # set the corresponding tensors into their correct target device.
+        target_device = inputs_embeds.device
+        batch_indices, non_image_indices, text_to_overwrite = (
+            batch_indices.to(target_device),
+            non_image_indices.to(target_device),
+            text_to_overwrite.to(target_device),
+        )
+        attention_mask = attention_mask.to(target_device)
+        input_ids = input_ids.to(target_device)
+
+        # 4. Fill the embeddings based on the mask. If we have ["hey" "<image>", "how", "are"]
+        # we need to index copy on [0, 577, 578, 579] for the text and [1:576] for the image features
+        final_embedding[batch_indices, text_to_overwrite] = inputs_embeds[batch_indices, non_image_indices]
+        final_attention_mask[batch_indices, text_to_overwrite] = attention_mask[batch_indices, non_image_indices]
+        final_input_ids[batch_indices, text_to_overwrite] = input_ids[batch_indices, non_image_indices]
+        final_labels = None
+        if labels is not None:
+            labels = labels.to(target_device)
+            final_labels = torch.full_like(final_attention_mask, ignore_index).to(torch.long)
+            final_labels[batch_indices, text_to_overwrite] = labels[batch_indices, non_image_indices]
+
+        # 5. Fill the embeddings corresponding to the images. Anything that is not `text_positions` needs filling (#29835)
+        with torch.no_grad():
+            image_to_overwrite = torch.full(
+                (batch_size, max_embed_dim), True, dtype=torch.bool, device=inputs_embeds.device
+            )
+            image_to_overwrite[batch_indices, text_to_overwrite] = False
+            embed_indices = torch.arange(max_embed_dim).unsqueeze(0).to(target_device)
+            embed_indices = embed_indices.expand(batch_size, max_embed_dim)
+            embed_seq_lens = embed_sequence_lengths[:, None].to(target_device)
+
+            if left_padding:
+                # exclude padding on the left
+                max_embed_dim = max_embed_dim.to(target_device)
+                val = (max_embed_dim - embed_indices) <= embed_seq_lens
+            else:
+                # exclude padding on the right
+                val = embed_indices < embed_seq_lens
+            image_to_overwrite &= val
+
+            if image_to_overwrite.sum() != num_image_features:
+                raise ValueError(
+                    f"{image_to_overwrite.sum()=} != {num_image_features=} The input provided to the model are wrong. "
+                    f"The number of image tokens is {torch.sum(special_image_token_mask)} while"
+                    f" the number of image given to the model is {num_images}. "
+                    f"This prevents correct indexing and breaks batch generation."
+                )
+        final_embedding[image_to_overwrite] = image_features.contiguous().reshape(-1, embed_dim).to(target_device)
+        final_attention_mask |= image_to_overwrite
+        position_ids = (final_attention_mask.cumsum(-1) - 1).masked_fill_((final_attention_mask == 0), 1)
+
+        return final_embedding, final_attention_mask, position_ids, final_labels, final_input_ids
+
+    def pack_image_features(self, image_features, image_sizes, vision_feature_select_strategy, image_newline=None):
+        """
+        Reshape, unpad and then pack each image_feature into a single image_features tensor containing all visual vectors.
+
+        Args:
+            image_features (`List[torch.Tensor]` of length num_images, each of shape `(num_patches, image_length, embed_dim)`)
+                List of image feature tensor, each contains all the visual feature of all patches.
+            image_sizes (`torch.Tensor` of shape `(num_images, 2)`)
+                Actual image size of each images (H, W).
+            vision_feature_select_strategy (`str`)
+                The feature selection strategy used to select the vision feature from the vision backbone.
+            image_newline (`torch.Tensor` of shape `(embed_dim)`)
+                New line embedding vector.
+        Returns:
+            image_features (`torch.Tensor` of shape `(all_feat_len, embed_dim)`)
+            feature_lens (`List[int]`)
+                token length of each image in image_features
+        """
+        new_image_features = []
+        feature_lens = []
+        for image_idx, image_feature in enumerate(image_features):
+            if image_feature.shape[0] > 1:
+                base_image_feature = image_feature[0]
+                image_feature = image_feature[1:]
+                height = width = self.config.vision_config.image_size // self.config.vision_config.patch_size
+
+                if vision_feature_select_strategy == "default":
+                    expected_num_patches = height * width
+                elif vision_feature_select_strategy == "full":
+                    expected_num_patches = height * width + 1
+                if expected_num_patches != base_image_feature.shape[0]:
+                    raise ValueError("The number of patches is not consistent with the image size.")
+
+                num_patch_height, num_patch_width = get_anyres_image_grid_shape(
+                    image_sizes[image_idx],
+                    self.config.image_grid_pinpoints,
+                    self.config.vision_config.image_size,
+                )
+                image_feature = image_feature.view(num_patch_height, num_patch_width, height, width, -1)
+                image_feature = image_feature.permute(4, 0, 2, 1, 3).contiguous()
+                image_feature = image_feature.flatten(1, 2).flatten(2, 3)
+                image_feature = unpad_image(image_feature, image_sizes[image_idx])
+                if image_newline is not None:
+                    image_feature = torch.cat(
+                        (
+                            image_feature,
+                            image_newline[:, None, None].expand(*image_feature.shape[:-1], 1).to(image_feature.dtype),
+                        ),
+                        dim=-1,
+                    )
+                image_feature = image_feature.flatten(1, 2).transpose(0, 1)
+                image_feature = torch.cat((base_image_feature, image_feature), dim=0)
+            else:
+                image_feature = image_feature[0]
+                if image_newline is not None:
+                    image_feature = torch.cat((image_feature, image_newline[None].to(image_feature)), dim=0)
+            new_image_features.append(image_feature)
+            feature_lens.append(image_feature.size(0))
+        image_features = torch.cat(new_image_features, dim=0)
+        feature_lens = torch.tensor(feature_lens, dtype=torch.long, device=image_features.device)
+        return image_features, feature_lens
+
+    def get_image_features(
+        self,
+        pixel_values: torch.FloatTensor,
+        image_sizes: torch.Tensor,
+        vision_feature_layer: int,
+        vision_feature_select_strategy: str,
+    ):
+        """
+        Obtains image last hidden states from the vision tower and apply multimodal projection.
+
+        Args:
+            pixel_values (`torch.FloatTensor]` of shape `(batch_size, num_patches, channels, height, width)`)
+               The tensors corresponding to the input images.
+            image_sizes (`torch.Tensor` of shape `(num_images, 2)`)
+                Actual image size of each images (H, W).
+            vision_feature_layer (`int`):
+                The index of the layer to select the vision feature.
+            vision_feature_select_strategy (`str`):
+                The feature selection strategy used to select the vision feature from the vision backbone.
+                Can be one of `"default"` or `"full"`
+        Returns:
+            image_features (List[`torch.Tensor`]): List of image feature tensor, each contains all the visual feature of all patches
+            and are of shape `(num_patches, image_length, embed_dim)`).
+        """
+        # ! infer image_num_patches from image_sizes
+        image_num_patches = [
+            image_size_to_num_patches(
+                image_size=imsize,
+                grid_pinpoints=self.config.image_grid_pinpoints,
+                patch_size=self.config.vision_config.image_size,
+            )
+            for imsize in image_sizes
+        ]
+        if pixel_values.dim() == 5:
+            # stacked if input is (batch_size, num_patches, num_channels, height, width)
+            _pixel_values_list = [pix_val[:num_patch] for pix_val, num_patch in zip(pixel_values, image_num_patches)]
+            pixel_values = torch.cat(_pixel_values_list, dim=0)
+        elif pixel_values.dim() != 4:
+            # otherwise has to be stacked from list of (num_patches, num_channels, height, width)
+            raise ValueError(f"pixel_values of shape {pixel_values.shape}, expect to be of 4 or 5 dimensions")
+
+        image_features = self.vision_tower(pixel_values, output_hidden_states=True)
+        selected_image_feature = image_features.hidden_states[vision_feature_layer]
+        if vision_feature_select_strategy == "default":
+        #     selected_image_feature = selected_image_feature[:, 1:]
+        # elif vision_feature_select_strategy == "full":
+            selected_image_feature = selected_image_feature
+        image_features, vq_loss = self.multi_modal_projector(selected_image_feature)
+        image_features = torch.split(image_features, image_num_patches, dim=0)
+        
+
+        return image_features, vq_loss
+
+    @add_start_docstrings_to_model_forward(LLAVA_NEXT_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=LlavaNextCausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        pixel_values: torch.FloatTensor = None,
+        image_sizes: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        vision_feature_layer: Optional[int] = None,
+        vision_feature_select_strategy: Optional[str] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        num_logits_to_keep: int = 0,
+    ) -> Union[Tuple, LlavaNextCausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+            num_logits_to_keep (`int`, *optional*):
+                Calculate logits for the last `num_logits_to_keep` tokens. If `0`, calculate logits for all
+                `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+                token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from PIL import Image
+        >>> import requests
+        >>> from transformers import AutoProcessor, LlavaNextForConditionalGeneration
+
+        >>> model = LlavaNextForConditionalGeneration.from_pretrained("llava-hf/llava-v1.6-mistral-7b-hf")
+        >>> processor = AutoProcessor.from_pretrained("llava-hf/llava-v1.6-mistral-7b-hf")
+
+        >>> prompt = "[INST] <image>\nWhat is shown in this image? [/INST]"
+        >>> url = "https://www.ilankelman.org/stopsigns/australia.jpg"
+        >>> image = Image.open(requests.get(url, stream=True).raw)
+
+        >>> inputs = processor(images=image, text=prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(**inputs, max_length=30)
+        >>> processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "[INST]  \nWhat is shown in this image? [/INST] The image appears to be a radar chart, which is a type of multi-dimensional plot (...)"
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+        vision_feature_layer = (
+            vision_feature_layer if vision_feature_layer is not None else self.config.vision_feature_layer
+        )
+        vision_feature_select_strategy = (
+            vision_feature_select_strategy
+            if vision_feature_select_strategy is not None
+            else self.config.vision_feature_select_strategy
+        )
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if pixel_values is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot specify both pixel_values and inputs_embeds at the same time, and must specify either one"
+            )
+
+        legacy_processing = False
+        if inputs_embeds is None:
+            inputs_embeds = self.get_input_embeddings()(input_ids)
+
+            # if the number of image tokens is more than image embeddings seq length, then prob we expanded it in processing
+            # not very reliable, but we don't expect one to actually pass 500+ images for one prompt
+            # In case we're in decoding stage, legacy behavior is checked by presence of pixel values even if use_cache=True
+            legacy_processing = (
+                (input_ids == self.config.image_token_index).sum(1).max() < self.config.image_seq_length
+            ) or (input_ids.shape[-1] == 1 and pixel_values is not None)
+
+        image_features = None
+        if pixel_values is not None and pixel_values.size(0) > 0:
+            image_features, vq_loss = self.get_image_features(
+                pixel_values,
+                image_sizes,
+                vision_feature_layer=vision_feature_layer,
+                vision_feature_select_strategy=vision_feature_select_strategy,
+            )
+
+            # NOTE we only support multimodal_patch_merge_type == "spatial_unpad"
+            image_features, feature_lens = self.pack_image_features(
+                image_features,
+                image_sizes,
+                vision_feature_select_strategy=vision_feature_select_strategy,
+                image_newline=self.image_newline,
+            )
+
+        if legacy_processing:
+            logger.warning_once(
+                "Expanding inputs for image tokens in LLaVa-NeXT should be done in processing. "
+                "Please add `patch_size` and `vision_feature_select_strategy` to the model's processing config or set directly "
+                "with `processor.patch_size = {{patch_size}}` and processor.vision_feature_select_strategy = {{vision_feature_select_strategy}}`. "
+                "Using processors without these attributes in the config is deprecated and will throw an error in v4.47."
+            )
+            if input_ids.shape[1] != 1:
+                inputs_embeds = inputs_embeds.to(image_features.dtype)
+                inputs_embeds, attention_mask, position_ids, labels, _ = self._merge_input_ids_with_image_features(
+                    image_features,
+                    feature_lens,
+                    inputs_embeds,
+                    input_ids,
+                    attention_mask,
+                    position_ids,
+                    labels=labels,
+                )
+                cache_position = torch.arange(attention_mask.shape[1], device=attention_mask.device)
+            else:
+                # Retrieve the first layer to inspect the logits and mask out the hidden states
+                # that are set to 0
+                first_layer_past_key_value = past_key_values[0][0][:, :, :, 0]
+
+                # Sum all dimensions of head_dim (-2) to avoid random errors such as: https://github.com/huggingface/transformers/pull/28032#issuecomment-1863691941
+                batch_index, non_attended_tokens = torch.where(first_layer_past_key_value.float().sum(-2) == 0)
+
+                # Get the target length
+                target_length = input_ids.shape[1]
+                past_length = first_layer_past_key_value.shape[-1]
+
+                extended_attention_mask = torch.ones(
+                    (attention_mask.shape[0], past_length),
+                    dtype=attention_mask.dtype,
+                    device=attention_mask.device,
+                )
+
+                # Filter out only the tokens that can be un-attended, this can happen
+                # if one uses Llava + Fused modules where the cache on the
+                # first iteration is already big enough, or if one passes custom cache
+                valid_indices = non_attended_tokens < extended_attention_mask.size(-1)
+                new_batch_index = batch_index[valid_indices]
+                new_non_attended_tokens = non_attended_tokens[valid_indices]
+
+                # Zero-out the places where we don't need to attend
+                extended_attention_mask[new_batch_index, new_non_attended_tokens] = 0
+                attention_mask = torch.cat((extended_attention_mask, attention_mask[:, -target_length:]), dim=1)
+                position_ids = torch.sum(attention_mask, dim=1).unsqueeze(-1) - 1
+                cache_position = torch.arange(attention_mask.shape[1], device=attention_mask.device)[-target_length:]
+
+        # TODO: @raushan retain only the new behavior after v4.47
+        elif image_features is not None:
+            n_image_tokens = (input_ids == self.config.image_token_index).sum().item()
+            n_image_features = image_features.shape[0]
+            if n_image_tokens != n_image_features:
+                raise ValueError(
+                    f"Image features and image tokens do not match: tokens: {n_image_tokens}, features {n_image_features}"
+                )
+            special_image_mask = (
+                (input_ids == self.config.image_token_index)
+                .unsqueeze(-1)
+                .expand_as(inputs_embeds)
+                .to(inputs_embeds.device)
+            )
+            image_features = image_features.to(inputs_embeds.device, inputs_embeds.dtype)
+            inputs_embeds = inputs_embeds.masked_scatter(special_image_mask, image_features)
+
+        outputs = self.language_model(
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            cache_position=cache_position,
+            num_logits_to_keep=num_logits_to_keep,
+        )
+
+        logits = outputs[0]
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            if attention_mask is not None:
+                # we use the input attention mask to shift the logits and labels, because it is 2D.
+                # we also crop attn mask in case it is longer, which happens in PrefixTuning with peft
+                shift_attention_mask = attention_mask[:, -(logits.shape[1] - 1) :].to(logits.device)
+                shift_logits = logits[..., :-1, :][shift_attention_mask.to(logits.device) != 0].contiguous()
+                shift_labels = labels[..., 1:][shift_attention_mask.to(labels.device) != 0].contiguous()
+            else:
+                shift_logits = logits[..., :-1, :].contiguous()
+                shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(
+                shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1).to(shift_logits.device)
+            )
+            print("This is original loss",loss)
+            #vq_loss = vq_loss.to(loss.device)
+            loss = loss 
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return LlavaNextCausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            image_hidden_states=image_features if pixel_values is not None else None,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        inputs_embeds=None,
+        pixel_values=None,
+        image_sizes=None,
+        attention_mask=None,
+        cache_position=None,
+        num_logits_to_keep=None,
+        **kwargs,
+    ):
+        # Overwritten -- in specific circumstances we don't want to forward image inputs to the model
+
+        model_inputs = self.language_model.prepare_inputs_for_generation(
+            input_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            num_logits_to_keep=num_logits_to_keep,
+            **kwargs,
+        )
+
+        # If we're in cached decoding stage, pixel values should be None because input ids do not contain special image token anymore
+        # Otherwise we need pixel values to be passed to model
+        if cache_position[0] == 0:
+            model_inputs["pixel_values"] = pixel_values
+            model_inputs["image_sizes"] = image_sizes
+
+        return model_inputs

test_on_new_qllava.ipynb

@@ -0,0 +1,2264 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from huggingface_hub import HfApi\n",
+    "from huggingface_hub import HfApi\n",
+    "api = HfApi()\n",
+    "# 替换为你的用户名/仓库名\n",
+    "repo_id = \"vincentchao/qllava-next\"\n",
+    "# # 创建仓库(如果还不存在)\n",
+    "api.create_repo(repo_id, repo_type=\"model\", private=False)\n",
+    "\n",
+    "# 上传整个目录\n",
+    "api.upload_folder(\n",
+    "    folder_path=\"/common/home/users/w/wzhao/vqclip/qllava_next_newest\",\n",
+    "    repo_id=repo_id,\n",
+    "    repo_type=\"model\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "classes2id = { 'neutral':0, 'porn':1,'gun':2,'cigarette':3,'alcohol':4, 'knife':5,'blood':6,'insulting_gesture':7}\n",
+    "id2class = ['neutral','porn','gun','cigarette','alcohol',\"knife\",'blood','insulting_gesture']"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "56dbb817e0a24197b2749a8ff82fe593",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Loading checkpoint shards:   0%|          | 0/7 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "from transformers import LlavaProcessor\n",
+    "import torch\n",
+    "import logging\n",
+    "from transformers import TrainerCallback\n",
+    "\n",
+    "import datetime\n",
+    "import os\n",
+    "#os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"  # 只显示第一个GPU设备\n",
+    "import torch  # 或 tensorflow 等\n",
+    "import torch.nn as nn\n",
+    "import os\n",
+    "import torch.distributed as dist\n",
+    "from transformers import LlavaProcessor\n",
+    "from qllava3_test import LlavaNextForConditionalGeneration\n",
+    "# %%\n",
+    "model_id = \"/common/public/llava/llama3-llava-next-8b-hf\"\n",
+    "processor = LlavaProcessor.from_pretrained(\n",
+    "    model_id ,\n",
+    ")\n",
+    "\n",
+    "# 加载模型\n",
+    "model = LlavaNextForConditionalGeneration.from_pretrained(\n",
+    "    \"/common/home/users/w/wzhao/vqclip/qllava_next_newest\", \n",
+    "    #\"/common/home/users/w/wzhao/qllava/vqllava/checkpoint-260\",\n",
+    "    # \"/common/home/users/w/wzhao/\"+path + \"/checkpoint-5000\",\n",
+    "    torch_dtype=torch.float32,  # 仍然使用float16以节省内存\n",
+    ").cuda()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Total images to process: 21213\n",
+      "Processing folder: /common/home/users/w/wzhao/llava_helper/nsfw_dataset_v1/porn/train (Category: porn)\n",
+      "this is q_latent_loss tensor(8.8170, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2042, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.6122, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9030, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.9424, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4856, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(11.4321, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.8580, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.0276, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0069, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.3328, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0832, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.4713, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8678, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.5524, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3881, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.4501, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6125, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.5718, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8930, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 10/21213 images\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.9837, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2459, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.0015, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2504, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(11.5371, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.8843, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.5971, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1493, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.3564, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5891, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.8574, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.7143, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.5497, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6374, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.1719, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2930, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.7135, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9284, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.7845, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9461, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 20/21213 images\n",
+      "this is q_latent_loss tensor(7.4899, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8725, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.1208, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0302, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.0984, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.7746, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.0674, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.7669, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(13.3643, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.3411, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.2910, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5727, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.9972, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2493, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.9211, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.7303, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.6993, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6748, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(6.4199, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.6050, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 30/21213 images\n",
+      "this is q_latent_loss tensor(9.1634, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2909, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.9688, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9922, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.6793, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4198, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.4200, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6050, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.4276, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6069, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.8010, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.7003, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.6418, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9105, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.3313, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0828, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.5423, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3856, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.8121, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2030, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 40/21213 images\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(12.9253, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.2313, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.9912, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2478, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(6.2949, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.5737, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(14.0637, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.5159, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(11.6149, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9037, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.6029, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.1507, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(13.2988, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.3247, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(13.8039, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.4510, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.4613, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1153, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(12.6695, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.1674, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 50/21213 images\n",
+      "this is q_latent_loss tensor(9.4005, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3501, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.8682, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.2171, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.6948, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1737, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.5272, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1318, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.7591, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1898, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.1201, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5300, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.5750, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6437, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.1127, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.7782, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.6827, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4207, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(11.4962, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.8740, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 60/21213 images\n",
+      "this is q_latent_loss tensor(9.2464, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3116, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.1726, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2932, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.0186, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.7546, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.8717, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4679, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(14.1204, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.5301, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.3599, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0900, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.3845, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.0961, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.8901, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9725, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.0965, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.0241, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.8583, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2146, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 70/21213 images\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.9591, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9898, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.9921, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2480, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.3415, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8354, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.7903, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4476, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.8003, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2001, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(6.3422, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.5856, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.4356, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1089, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.7969, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4492, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.9973, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.7493, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.3360, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0840, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 80/21213 images\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.2751, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.8188, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.1294, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0324, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.8635, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9659, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.8551, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9638, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.8822, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4705, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.0291, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0073, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.0501, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2625, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(12.6947, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.1737, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.9940, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4985, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.2904, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8226, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 90/21213 images\n",
+      "this is q_latent_loss tensor(9.6648, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4162, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.5542, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3886, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.8552, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4638, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.3666, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0916, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.7352, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4338, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.1629, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0407, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(14.1190, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.5297, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.3341, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.8335, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.9301, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2325, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(6.2145, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.5536, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 100/21213 images\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.1991, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0498, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.6093, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9023, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.7087, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4272, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.8062, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2016, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.4319, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1080, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.5253, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1313, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.1614, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0404, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.9642, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4910, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.2686, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3171, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.1140, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5285, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 110/21213 images\n",
+      "this is q_latent_loss tensor(12.3632, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.0908, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.1329, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0332, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.9918, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2479, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.1709, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.7927, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.1411, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5353, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.4103, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6026, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.9719, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4930, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(6.5009, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.6252, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.8178, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9544, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.2491, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.0623, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 120/21213 images\n",
+      "this is q_latent_loss tensor(11.7427, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.9357, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.0834, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0208, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.1786, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.0447, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.7889, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4472, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.3587, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0897, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.1803, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0451, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.9105, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4776, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.8691, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2173, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.4474, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8619, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.4359, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.3590, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 130/21213 images\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(6.5443, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.6361, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.5185, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.8796, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(15.6529, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.9132, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(11.5756, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.8939, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.6367, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.1592, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.2915, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5729, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.9234, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4809, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(12.2591, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.0648, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.1375, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2844, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.9599, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2400, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 140/21213 images\n",
+      "this is q_latent_loss tensor(10.5039, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6260, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.1007, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2752, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(13.8705, device='cuda:0')\n",
+      "This is e_latent_loss tensor(3.4676, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.4230, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.6058, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.0464, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5116, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(10.8465, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.7116, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.8604, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4651, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(7.0887, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.7722, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(9.0656, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.2664, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.3511, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0878, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "Processed 150/21213 images\n",
+      "this is q_latent_loss tensor(11.1818, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.7954, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(7.8044, device='cuda:0')\n",
+      "This is e_latent_loss tensor(1.9511, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.0274, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0068, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(8.4584, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.1146, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "this is q_latent_loss tensor(8.1515, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.0379, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(9.8069, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.4517, device='cuda:0')\n",
+      "57\n",
+      "[0]\n",
+      "this is q_latent_loss tensor(10.3290, device='cuda:0')\n",
+      "This is e_latent_loss tensor(2.5823, device='cuda:0')\n",
+      "57\n",
+      "[0]\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n",
+      "Setting `pad_token_id` to `eos_token_id`:None for open-end generation.\n"
+     ]
+    }
+   ],
+   "source": [
+    "from PIL import Image\n",
+    "import torch\n",
+    "import os\n",
+    "import json\n",
+    "import re\n",
+    "\n",
+    "# 假设 processor 和 model 已经初始化好了\n",
+    "# 如果没有，需要添加相关的初始化代码\n",
+    "\n",
+    "def get_category_from_path(path):\n",
+    "    \"\"\"从路径中提取类别\"\"\"\n",
+    "    if \"porn\" in path:\n",
+    "        return \"porn\"\n",
+    "    elif \"blood\" in path:\n",
+    "        return \"blood\"\n",
+    "    elif \"gun\" in path:\n",
+    "        return \"gun\"\n",
+    "    elif \"insulting_gesture\" in path:\n",
+    "        return \"insulting_gesture\"\n",
+    "    elif \"alcohol\" in path:\n",
+    "        return \"alcohol\"\n",
+    "    elif \"cigarette\" in path:\n",
+    "        return \"cigarette\"\n",
+    "    elif \"knife\" in path:\n",
+    "        return \"knife\"\n",
+    "    elif \"real_neutral\" in path:\n",
+    "        return \"neutral\"\n",
+    "    else:\n",
+    "        return \"unknown\"\n",
+    "\n",
+    "def process_image(image_path, processor, model):\n",
+    "    \"\"\"处理图像并获取描述\"\"\"\n",
+    "    try:\n",
+    "        image = Image.open(image_path)\n",
+    "        \n",
+    "        conversation = [\n",
+    "            {\n",
+    "              \"role\": \"user\",\n",
+    "              \"content\": [\n",
+    "                  {\"type\": \"text\", \"text\": \"Describe this image \"},\n",
+    "                  {\"type\": \"image\"},\n",
+    "                ],\n",
+    "            }, \n",
+    "        ]\n",
+    "        \n",
+    "        prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)\n",
+    "        \n",
+    "        inputs = processor(images=image, text=prompt, return_tensors=\"pt\").to(model.device)\n",
+    "        \n",
+    "        with torch.no_grad():\n",
+    "            output = model.generate(**inputs, max_new_tokens=10)  # 增加 token 数量以获取更详细的描述\n",
+    "        \n",
+    "        description = processor.decode(output[0], skip_special_tokens=True)\n",
+    "        \n",
+    "        # 清理描述文本，移除提示和可能的系统消息\n",
+    "        # 这部分可能需要根据实际输出进行调整\n",
+    "        description = re.sub(r\"Describe this image\\s*\", \"\", description)\n",
+    "        \n",
+    "        return description\n",
+    "    except ValueError as e:\n",
+    "        error_data = e.args[0]  # 这将获取到 [indices, categories[0]]\n",
+    "        \n",
+    "        # 提取 indices 和 categories[0]\n",
+    "        if isinstance(error_data, list) and len(error_data) == 2:\n",
+    "            captured_indices = error_data[0]\n",
+    "            captured_category = error_data[1]\n",
+    "\n",
+    "        return f\"Error processing image: {id2class[captured_category]} index {captured_indices}\"\n",
+    "\n",
+    "def main():\n",
+    "    image_folders = [\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/nsfw_dataset_v1/porn/train\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/nsfw_dataset_v1/porn/test\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/blood/testing/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/blood/training/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/gun/testing/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/gun/training/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/insulting_gesture/training/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/insulting_gesture/testing/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/alcohol/testing/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/alcohol/training/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/cigarette/testing/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/cigarette/training/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/knife/testing/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/dataset/class/knife/training/jpg\",\n",
+    "        \"/common/home/users/w/wzhao/llava_helper/real_neutral\"\n",
+    "    ]\n",
+    "    \n",
+    "    results = []\n",
+    "    total_images = 0\n",
+    "    \n",
+    "    # 计算总图片数量\n",
+    "    for folder in image_folders:\n",
+    "        if os.path.exists(folder):\n",
+    "            for root, _, files in os.walk(folder):\n",
+    "                total_images += sum(1 for f in files if f.lower().endswith(('.png', '.jpg', '.jpeg')))\n",
+    "    \n",
+    "    print(f\"Total images to process: {total_images}\")\n",
+    "    \n",
+    "    processed_count = 0\n",
+    "    \n",
+    "    # 处理每个文件夹中的图片\n",
+    "    for folder in image_folders:\n",
+    "        if not os.path.exists(folder):\n",
+    "            print(f\"Folder not found: {folder}\")\n",
+    "            continue\n",
+    "            \n",
+    "        category = get_category_from_path(folder)\n",
+    "        print(f\"Processing folder: {folder} (Category: {category})\")\n",
+    "        \n",
+    "        for root, _, files in os.walk(folder):\n",
+    "            for file in files:\n",
+    "                if file.lower().endswith(('.png', '.jpg', '.jpeg')):\n",
+    "                    image_path = os.path.join(root, file)\n",
+    "                    \n",
+    "                    # 处理图像\n",
+    "                    description = process_image(image_path, processor, model)\n",
+    "                    \n",
+    "                    # 存储结果\n",
+    "                    results.append({\n",
+    "                        \"path\": image_path,\n",
+    "                        \"response\": description,\n",
+    "                        \"category\": category\n",
+    "                    })\n",
+    "                    \n",
+    "                    processed_count += 1\n",
+    "                    if processed_count % 10 == 0:\n",
+    "                        print(f\"Processed {processed_count}/{total_images} images\")\n",
+    "                    \n",
+    "                    # 每处理100张图片保存一次中间结果\n",
+    "                    if processed_count % 100 == 0:\n",
+    "                        with open(f\"/common/home/users/w/wzhao/vqclip/json_results_newest/image_descriptions_checkpoint_{processed_count}.json\", \"w\") as f:\n",
+    "                            json.dump(results, f, indent=2)\n",
+    "    \n",
+    "    # 保存最终结果\n",
+    "    with open(\"/common/home/users/w/wzhao/vqclip/json_results_newest/image_descriptions_complete.json\", \"w\") as f:\n",
+    "        json.dump(results, f, indent=2)\n",
+    "    \n",
+    "    print(f\"Processing complete. Results saved to image_descriptions_complete.json\")\n",
+    "\n",
+    "if __name__ == \"__main__\":\n",
+    "    # 确保已经初始化了processor和model\n",
+    "    # 如果没有，需要在这里添加初始化代码\n",
+    "    main()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "array(['54', 'porn'], dtype='<U21')"
+      ]
+     },
+     "execution_count": 4,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import numpy as np\n",
+    "np.load(\"/common/home/users/w/wzhao/vqclip/classified_results/1.npy\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "**************************************************************\n",
+      "torch.Size([1, 3, 336, 336])\n",
+      "this is q_latent_loss tensor(0.6994, device='cuda:0')\n",
+      "This is e_latent_loss tensor(0.1748, device='cuda:0')\n",
+      "34\n",
+      "[1]\n",
+      "1\n",
+      "检测到不适当内容，生成已停止\n",
+      "**************************************************************\n"
+     ]
+    }
+   ],
+   "source": [
+    "from PIL import Image\n",
+    "import time\n",
+    "image = Image.open('/common/home/users/w/wzhao/qllava/00008.jpg')\n",
+    "#image = Image.open('/common/home/users/w/wzhao/qllava/94915.jpg')\n",
+    "#image = Image.open(\"\")\n",
+    "# Define a chat histiry and use `apply_chat_template` to get correctly formatted prompt\n",
+    "# Each value in \"content\" has to be a list of dicts with types (\"text\", \"image\") \n",
+    "print(\"**************************************************************\")\n",
+    "\n",
+    "\n",
+    "conversation = [\n",
+    "    {\n",
+    "      \"role\": \"user\",\n",
+    "      \"content\": [\n",
+    "          {\"type\": \"text\", \"text\": \"Describe this image \"},\n",
+    "          {\"type\": \"image\"},\n",
+    "        ],\n",
+    "    }, \n",
+    "]\n",
+    "\n",
+    "prompt = processor.apply_chat_template(conversation, add_generation_prompt=True)\n",
+    "#print(prompt)\n",
+    "\n",
+    "inputs = processor(images=image, text=prompt, return_tensors=\"pt\").to(model.device)\n",
+    "print(inputs[\"pixel_values\"].shape)\n",
+    "\n",
+    "# autoregressively complete prompt\n",
+    "try:\n",
+    "    output = model.generate(**inputs, max_new_tokens=50)\n",
+    "    print(processor.decode(output[0], skip_special_tokens=True))\n",
+    "except ValueError as e:\n",
+    "    # 记录错误信息\n",
+    "    \n",
+    "    harm_class = int(str(e)))\n",
+    "\n",
+    "    \n",
+    "    # 可以选择继续执行其他代码或者返回一个预设的安全输出\n",
+    "    print(\"检测到不适当内容，生成已停止\")\n",
+    "print(\"**************************************************************\")\n",
+    "#time.sleep(100)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "LlavaForConditionalGeneration(\n",
+       "  (vision_tower): CLIPVisionModel(\n",
+       "    (vision_model): CLIPVisionTransformer(\n",
+       "      (embeddings): CLIPVisionEmbeddings(\n",
+       "        (patch_embedding): Conv2d(3, 1024, kernel_size=(14, 14), stride=(14, 14), bias=False)\n",
+       "        (position_embedding): Embedding(577, 1024)\n",
+       "      )\n",
+       "      (pre_layrnorm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
+       "      (encoder): CLIPEncoder(\n",
+       "        (layers): ModuleList(\n",
+       "          (0-23): 24 x CLIPEncoderLayer(\n",
+       "            (self_attn): CLIPSdpaAttention(\n",
+       "              (k_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
+       "              (v_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
+       "              (q_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
+       "              (out_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
+       "            )\n",
+       "            (layer_norm1): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
+       "            (mlp): CLIPMLP(\n",
+       "              (activation_fn): QuickGELUActivation()\n",
+       "              (fc1): Linear(in_features=1024, out_features=4096, bias=True)\n",
+       "              (fc2): Linear(in_features=4096, out_features=1024, bias=True)\n",
+       "            )\n",
+       "            (layer_norm2): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
+       "          )\n",
+       "        )\n",
+       "      )\n",
+       "      (post_layernorm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
+       "    )\n",
+       "  )\n",
+       "  (multi_modal_projector): LlavaMultiModalProjector(\n",
+       "    (linear_1): Linear(in_features=1024, out_features=4096, bias=True)\n",
+       "    (act): GELUActivation()\n",
+       "    (linear_2): Linear(in_features=4096, out_features=4096, bias=True)\n",
+       "    (vq): VectorQuantizer(\n",
+       "      (embedding): Embedding(16000, 4096)\n",
+       "    )\n",
+       "    (vq_cls): VectorQuantizerCLS(\n",
+       "      (embedding): Embedding(128, 4096)\n",
+       "    )\n",
+       "  )\n",
+       "  (language_model): LlamaForCausalLM(\n",
+       "    (model): LlamaModel(\n",
+       "      (embed_tokens): Embedding(32064, 4096)\n",
+       "      (layers): ModuleList(\n",
+       "        (0-31): 32 x LlamaDecoderLayer(\n",
+       "          (self_attn): LlamaSdpaAttention(\n",
+       "            (q_proj): Linear(in_features=4096, out_features=4096, bias=False)\n",
+       "            (k_proj): Linear(in_features=4096, out_features=4096, bias=False)\n",
+       "            (v_proj): Linear(in_features=4096, out_features=4096, bias=False)\n",
+       "            (o_proj): Linear(in_features=4096, out_features=4096, bias=False)\n",
+       "            (rotary_emb): LlamaRotaryEmbedding()\n",
+       "          )\n",
+       "          (mlp): LlamaMLP(\n",
+       "            (gate_proj): Linear(in_features=4096, out_features=11008, bias=False)\n",
+       "            (up_proj): Linear(in_features=4096, out_features=11008, bias=False)\n",
+       "            (down_proj): Linear(in_features=11008, out_features=4096, bias=False)\n",
+       "            (act_fn): SiLU()\n",
+       "          )\n",
+       "          (input_layernorm): LlamaRMSNorm((4096,), eps=1e-05)\n",
+       "          (post_attention_layernorm): LlamaRMSNorm((4096,), eps=1e-05)\n",
+       "        )\n",
+       "      )\n",
+       "      (norm): LlamaRMSNorm((4096,), eps=1e-05)\n",
+       "      (rotary_emb): LlamaRotaryEmbedding()\n",
+       "    )\n",
+       "    (lm_head): Linear(in_features=4096, out_features=32064, bias=False)\n",
+       "  )\n",
+       ")"
+      ]
+     },
+     "execution_count": 10,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[2025-03-28 01:00:17,048] [WARNING] [real_accelerator.py:174:get_accelerator] Setting accelerator to CPU. If you have GPU or other accelerator, we were unable to detect it.\n",
+      "[2025-03-28 01:00:17,049] [INFO] [real_accelerator.py:219:get_accelerator] Setting ds_accelerator to cpu (auto detect)\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/opt/apps/software/Anaconda3/2024.06-1/compiler_compat/ld: cannot find -laio: No such file or directory\n",
+      "collect2: error: ld returned 1 exit status\n"
+     ]
+    },
+    {
+     "ename": "AttributeError",
+     "evalue": "'list' object has no attribute 'device'",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mAttributeError\u001b[0m                            Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[6], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[43mmodel\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43msave_pretrained\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m      2\u001b[0m \u001b[43m    \u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43m/common/home/users/w/wzhao/vqclip/VQLLMfinal\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\n\u001b[1;32m      3\u001b[0m \u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m~/.local/lib/python3.12/site-packages/transformers/modeling_utils.py:2971\u001b[0m, in \u001b[0;36mPreTrainedModel.save_pretrained\u001b[0;34m(self, save_directory, is_main_process, state_dict, save_function, push_to_hub, max_shard_size, safe_serialization, variant, token, save_peft_format, **kwargs)\u001b[0m\n\u001b[1;32m   2968\u001b[0m     weights_name \u001b[38;5;241m=\u001b[39m ADAPTER_SAFE_WEIGHTS_NAME \u001b[38;5;28;01mif\u001b[39;00m safe_serialization \u001b[38;5;28;01melse\u001b[39;00m ADAPTER_WEIGHTS_NAME\n\u001b[1;32m   2970\u001b[0m filename_pattern \u001b[38;5;241m=\u001b[39m weights_name\u001b[38;5;241m.\u001b[39mreplace(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m.bin\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{suffix}\u001b[39;00m\u001b[38;5;124m.bin\u001b[39m\u001b[38;5;124m\"\u001b[39m)\u001b[38;5;241m.\u001b[39mreplace(\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m.safetensors\u001b[39m\u001b[38;5;124m\"\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{suffix}\u001b[39;00m\u001b[38;5;124m.safetensors\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[0;32m-> 2971\u001b[0m state_dict_split \u001b[38;5;241m=\u001b[39m \u001b[43msplit_torch_state_dict_into_shards\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m   2972\u001b[0m \u001b[43m    \u001b[49m\u001b[43mstate_dict\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mfilename_pattern\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mfilename_pattern\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mmax_shard_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmax_shard_size\u001b[49m\n\u001b[1;32m   2973\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   2974\u001b[0m \u001b[38;5;66;03m# Save index if sharded\u001b[39;00m\n\u001b[1;32m   2975\u001b[0m index \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
+      "File \u001b[0;32m~/.local/lib/python3.12/site-packages/huggingface_hub/serialization/_torch.py:369\u001b[0m, in \u001b[0;36msplit_torch_state_dict_into_shards\u001b[0;34m(state_dict, filename_pattern, max_shard_size)\u001b[0m\n\u001b[1;32m    302\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21msplit_torch_state_dict_into_shards\u001b[39m(\n\u001b[1;32m    303\u001b[0m     state_dict: Dict[\u001b[38;5;28mstr\u001b[39m, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtorch.Tensor\u001b[39m\u001b[38;5;124m\"\u001b[39m],\n\u001b[1;32m    304\u001b[0m     \u001b[38;5;241m*\u001b[39m,\n\u001b[1;32m    305\u001b[0m     filename_pattern: \u001b[38;5;28mstr\u001b[39m \u001b[38;5;241m=\u001b[39m constants\u001b[38;5;241m.\u001b[39mSAFETENSORS_WEIGHTS_FILE_PATTERN,\n\u001b[1;32m    306\u001b[0m     max_shard_size: Union[\u001b[38;5;28mint\u001b[39m, \u001b[38;5;28mstr\u001b[39m] \u001b[38;5;241m=\u001b[39m MAX_SHARD_SIZE,\n\u001b[1;32m    307\u001b[0m ) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m StateDictSplit:\n\u001b[1;32m    308\u001b[0m \u001b[38;5;250m    \u001b[39m\u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m    309\u001b[0m \u001b[38;5;124;03m    Split a model state dictionary in shards so that each shard is smaller than a given size.\u001b[39;00m\n\u001b[1;32m    310\u001b[0m \n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    367\u001b[0m \u001b[38;5;124;03m    ```\u001b[39;00m\n\u001b[1;32m    368\u001b[0m \u001b[38;5;124;03m    \"\"\"\u001b[39;00m\n\u001b[0;32m--> 369\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43msplit_state_dict_into_shards_factory\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    370\u001b[0m \u001b[43m        \u001b[49m\u001b[43mstate_dict\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    371\u001b[0m \u001b[43m        \u001b[49m\u001b[43mmax_shard_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmax_shard_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    372\u001b[0m \u001b[43m        \u001b[49m\u001b[43mfilename_pattern\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mfilename_pattern\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    373\u001b[0m \u001b[43m        \u001b[49m\u001b[43mget_storage_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mget_torch_storage_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    374\u001b[0m \u001b[43m        \u001b[49m\u001b[43mget_storage_id\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mget_torch_storage_id\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    375\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m~/.local/lib/python3.12/site-packages/huggingface_hub/serialization/_base.py:108\u001b[0m, in \u001b[0;36msplit_state_dict_into_shards_factory\u001b[0;34m(state_dict, get_storage_size, filename_pattern, get_storage_id, max_shard_size)\u001b[0m\n\u001b[1;32m    105\u001b[0m     \u001b[38;5;28;01mcontinue\u001b[39;00m\n\u001b[1;32m    107\u001b[0m \u001b[38;5;66;03m# If a `tensor` shares the same underlying storage as another tensor, we put `tensor` in the same `block`\u001b[39;00m\n\u001b[0;32m--> 108\u001b[0m storage_id \u001b[38;5;241m=\u001b[39m \u001b[43mget_storage_id\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtensor\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    109\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m storage_id \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m    110\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m storage_id \u001b[38;5;129;01min\u001b[39;00m storage_id_to_tensors:\n\u001b[1;32m    111\u001b[0m         \u001b[38;5;66;03m# We skip this tensor for now and will reassign to correct shard later\u001b[39;00m\n",
+      "File \u001b[0;32m~/.local/lib/python3.12/site-packages/huggingface_hub/serialization/_torch.py:746\u001b[0m, in \u001b[0;36mget_torch_storage_id\u001b[0;34m(tensor)\u001b[0m\n\u001b[1;32m    735\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mget_torch_storage_id\u001b[39m(tensor: \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtorch.Tensor\u001b[39m\u001b[38;5;124m\"\u001b[39m) \u001b[38;5;241m-\u001b[39m\u001b[38;5;241m>\u001b[39m Optional[Tuple[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtorch.device\u001b[39m\u001b[38;5;124m\"\u001b[39m, Union[\u001b[38;5;28mint\u001b[39m, Tuple[Any, \u001b[38;5;241m.\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;241m.\u001b[39m]], \u001b[38;5;28mint\u001b[39m]]:\n\u001b[1;32m    736\u001b[0m \u001b[38;5;250m    \u001b[39m\u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m    737\u001b[0m \u001b[38;5;124;03m    Return unique identifier to a tensor storage.\u001b[39;00m\n\u001b[1;32m    738\u001b[0m \n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    744\u001b[0m \u001b[38;5;124;03m    Taken from https://github.com/huggingface/transformers/blob/1ecf5f7c982d761b4daaa96719d162c324187c64/src/transformers/pytorch_utils.py#L278.\u001b[39;00m\n\u001b[1;32m    745\u001b[0m \u001b[38;5;124;03m    \"\"\"\u001b[39;00m\n\u001b[0;32m--> 746\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[43mtensor\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdevice\u001b[49m\u001b[38;5;241m.\u001b[39mtype \u001b[38;5;241m==\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mmeta\u001b[39m\u001b[38;5;124m\"\u001b[39m:\n\u001b[1;32m    747\u001b[0m         \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    748\u001b[0m     \u001b[38;5;28;01melse\u001b[39;00m:\n",
+      "\u001b[0;31mAttributeError\u001b[0m: 'list' object has no attribute 'device'"
+     ]
+    }
+   ],
+   "source": [
+    "model.save_pretrained(\n",
+    "    \"/common/home/users/w/wzhao/vqclip/VQLLMfinal\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "<class 'qllava_new.LlavaForConditionalGeneration'>\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(type(model))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "array([56,  1])"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import numpy as np\n",
+    "np.load(\"/common/home/users/w/wzhao/vqclip/classified_results_llama2/1.npy\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "统计结果已保存到字典并导出到：\n",
+      "- JSON: /common/home/users/w/wzhao/vqclip/classification_stats.json\n",
+      "- Pickle: /common/home/users/w/wzhao/vqclip/classification_stats.pkl\n",
+      "\n",
+      "基本统计信息:\n",
+      "总共发现 8 个不同的类别\n",
+      "总共处理了 21213 个元素\n",
+      "\n",
+      "各类别的统计摘要:\n",
+      "类别 0: 包含 5413 个元素，有 43 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 16: 615次, ID 60: 609次, ID 14: 566次\n",
+      "类别 1: 包含 5554 个元素，有 4 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 56: 2067次, ID 7: 1430次, ID 34: 1180次\n",
+      "类别 2: 包含 1473 个元素，有 3 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 13: 778次, ID 24: 488次, ID 23: 207次\n",
+      "类别 3: 包含 2134 个元素，有 4 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 46: 840次, ID 33: 641次, ID 39: 351次\n",
+      "类别 4: 包含 1416 个元素，有 2 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 22: 748次, ID 42: 668次\n",
+      "类别 5: 包含 2785 个元素，有 2 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 52: 1654次, ID 4: 1131次\n",
+      "类别 6: 包含 1723 个元素，有 5 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 20: 555次, ID 57: 481次, ID 17: 350次\n",
+      "类别 7: 包含 715 个元素，有 1 个不同的元素ID\n",
+      "  出现频率最高的元素: ID 40: 715次\n"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "import numpy as np\n",
+    "import json\n",
+    "from collections import defaultdict\n",
+    "import pickle\n",
+    "\n",
+    "# 定义目录路径\n",
+    "directory_path = '/common/home/users/w/wzhao/vqclip/classified_results_llama2'\n",
+    "\n",
+    "# 创建一个嵌套字典，用于存储每个类别中每个元素ID出现的次数\n",
+    "class_element_counts = defaultdict(lambda: defaultdict(int))\n",
+    "# 创建一个字典用于存储每个类别的总计数\n",
+    "class_total_counts = defaultdict(int)\n",
+    "\n",
+    "# 遍历目录中的所有.npy文件\n",
+    "try:\n",
+    "    for filename in os.listdir(directory_path):\n",
+    "        if filename.endswith('.npy'):\n",
+    "            file_path = os.path.join(directory_path, filename)\n",
+    "            \n",
+    "            # 加载.npy文件\n",
+    "            data = np.load(file_path)\n",
+    "            \n",
+    "            # 确保数据格式正确\n",
+    "            if data.size == 2:\n",
+    "                element_id = int(data[0])  # 确保是整数\n",
+    "                class_id = int(data[1])    # 确保是整数\n",
+    "                \n",
+    "                # 增加该元素在对应类别中的计数\n",
+    "                class_element_counts[class_id][element_id] += 1\n",
+    "                # 增加该类别的总计数\n",
+    "                class_total_counts[class_id] += 1\n",
+    "            else:\n",
+    "                print(f\"警告: 文件 {filename} 的数据格式不符合预期，已跳过\")\n",
+    "    \n",
+    "    # 将defaultdict转换为普通dict以便序列化\n",
+    "    result_dict = {\n",
+    "        \"class_totals\": dict(class_total_counts),\n",
+    "        \"class_elements\": {\n",
+    "            class_id: dict(elements) \n",
+    "            for class_id, elements in class_element_counts.items()\n",
+    "        }\n",
+    "    }\n",
+    "    \n",
+    "    # 保存结果到JSON文件\n",
+    "    output_json_path = os.path.join(os.path.dirname(directory_path), \"classification_stats.json\")\n",
+    "    with open(output_json_path, 'w') as f:\n",
+    "        json.dump(result_dict, f, indent=2)\n",
+    "    \n",
+    "    # 也保存为Python pickle格式，这样在后续Python处理中更方便\n",
+    "    output_pickle_path = os.path.join(os.path.dirname(directory_path), \"classification_stats.pkl\")\n",
+    "    with open(output_pickle_path, 'wb') as f:\n",
+    "        pickle.dump(result_dict, f)\n",
+    "    \n",
+    "    # 打印一些基本统计信息\n",
+    "    print(f\"统计结果已保存到字典并导出到：\")\n",
+    "    print(f\"- JSON: {output_json_path}\")\n",
+    "    print(f\"- Pickle: {output_pickle_path}\")\n",
+    "    print(\"\\n基本统计信息:\")\n",
+    "    print(f\"总共发现 {len(class_total_counts)} 个不同的类别\")\n",
+    "    total_elements = sum(class_total_counts.values())\n",
+    "    print(f\"总共处理了 {total_elements} 个元素\")\n",
+    "    \n",
+    "    # 打印每个类别的样本统计\n",
+    "    print(\"\\n各类别的统计摘要:\")\n",
+    "    for class_id in sorted(class_total_counts.keys()):\n",
+    "        print(f\"类别 {class_id}: 包含 {class_total_counts[class_id]} 个元素，有 {len(class_element_counts[class_id])} 个不同的元素ID\")\n",
+    "        \n",
+    "        # 获取该类别中出现次数最多的3个元素\n",
+    "        sorted_elements = sorted(class_element_counts[class_id].items(), \n",
+    "                                key=lambda x: x[1], reverse=True)[:3]\n",
+    "        \n",
+    "        # 打印这些元素及其出现次数\n",
+    "        print(f\"  出现频率最高的元素: \" + \", \".join([f\"ID {e_id}: {count}次\" for e_id, count in sorted_elements]))\n",
+    "\n",
+    "except Exception as e:\n",
+    "    print(f\"发生错误: {e}\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.4"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}