---
library_name: transformers
license: apache-2.0
base_model: IntervitensInc/Mistral-Nemo-Base-2407-chatml
tags:
- axolotl
- generated_from_trainer
datasets:
- lightblue/gpt4_conversations_multilingual
- Nopm/Opus_WritingStruct
model-index:
- name: hiwaifu-12b
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.8.1`
```yaml
base_model: IntervitensInc/Mistral-Nemo-Base-2407-chatml
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: taozi555/hiwaifu-12b
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: /root/taozi555/deepseek-rp/model12_digg1_safe.jsonl
    conversation: chatml
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: /root/taozi555/deepseek-rp/model12_digg1_unsafe.jsonl
    conversation: chatml
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: /root/taozi555/deepseek-rp/model2_digg1_unsafe.jsonl
    conversation: chatml
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: /root/taozi555/deepseek-rp/model2_digg1_safe.jsonl
    conversation: chatml
    type: chat_template
    field_messages: conversations
    message_field_role: role
    message_field_content: content
  - path: /root/processed_SCP_40k_dataset
    conversation: chatml
    type: chat_template
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    split: train
  - path: lightblue/gpt4_conversations_multilingual
    conversation: chatml
    type: chat_template
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    split: train
  - path: Nopm/Opus_WritingStruct
    conversation: chatml
    type: chat_template
    #field_messages: messages
    message_field_role: role
    message_field_content: content
    split: train
#  - path: Gryphe/Sonnet3.5-SlimOrcaDedupCleaned
#    conversation: chatml
#    type: chat_template
#    field_messages: conversations
#    message_field_role: from
#    message_field_content: value
#    split: train
chat_template: chatml
shuffle_merged_datasets: true
default_system_message: "You are the JadeSpeech model from the HiWaifu App."
dataset_prepared_path: /root/autodl-tmp/data/
val_set_size: 0.05
output_dir: /root/autodl-tmp/hiwaifu-12b/

sequence_len: 32768
sample_packing: true
pad_to_sequence_len: true

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:
lora_fan_in_fan_out:

wandb_project: hiwaifu-12b-v4
wandb_entity:
wandb_watch:
wandb_name: hiwaifu-12b-v4
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00005

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint: /root/autodl-tmp/hiwaifu-12b/checkpoint-1030
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 40
evals_per_epoch:
eval_table_size:
eval_max_new_tokens:
saves_per_epoch: 2
debug:
deepspeed: /root/deepspeed_configs/zero3_bf16.json
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  pad_token: <pad>
```

</details><br>

# hiwaifu-12b

This model is a fine-tuned version of [IntervitensInc/Mistral-Nemo-Base-2407-chatml](https://huggingface.co/IntervitensInc/Mistral-Nemo-Base-2407-chatml) on the /root/taozi555/deepseek-rp/model12_digg1_safe.jsonl, the /root/taozi555/deepseek-rp/model12_digg1_unsafe.jsonl, the /root/taozi555/deepseek-rp/model2_digg1_unsafe.jsonl, the /root/taozi555/deepseek-rp/model2_digg1_safe.jsonl, the lightblue/gpt4_conversations_multilingual and the Nopm/Opus_WritingStruct datasets.
It achieves the following results on the evaluation set:
- Loss: 0.1903

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- num_devices: 5
- gradient_accumulation_steps: 2
- total_train_batch_size: 10
- total_eval_batch_size: 5
- optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 40
- num_epochs: 2.0

### Training results

| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 0.1899        | 2.0   | 1466 | 0.1903          |


### Framework versions

- Transformers 4.51.0
- Pytorch 2.5.1+cu124
- Datasets 3.5.0
- Tokenizers 0.21.1