yujiepan/qwq-tiny-random-dim64

This tiny model is for debugging. It is randomly initialized with the config adapted from Qwen/QwQ-32B.

Example usage:

from transformers import pipeline
model_id = "yujiepan/qwq-tiny-random-dim64"
pipe = pipeline(
    "text-generation", model=model_id, device="cuda",
    trust_remote_code=True, max_new_tokens=3,
)
print(pipe("Hello World!"))

Codes to create this repo:

import torch

from transformers import (
    AutoConfig,
    AutoModelForCausalLM,
    AutoTokenizer,
    GenerationConfig,
    pipeline,
    set_seed,
)

source_model_id = "Qwen/QwQ-32B"
save_folder = "/tmp/yujiepan/qwq-tiny-random-dim64"

tokenizer = AutoTokenizer.from_pretrained(
    source_model_id, trust_remote_code=True,
)
tokenizer.save_pretrained(save_folder)

config = AutoConfig.from_pretrained(
    source_model_id, trust_remote_code=True,
)
config._name_or_path = source_model_id
config.hidden_size = 64
config.intermediate_size = 64
config.num_key_value_heads = 1
config.num_attention_heads = 2
config.num_hidden_layers = 2
config.max_window_layers = 1
model = AutoModelForCausalLM.from_config(
    config,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)
model.generation_config = GenerationConfig.from_pretrained(
    source_model_id, trust_remote_code=True,
)
set_seed(42)
with torch.no_grad():
    for name, p in sorted(model.named_parameters()):
        torch.nn.init.normal_(p, 0, 0.5)
        print(name, p.shape)
model.save_pretrained(save_folder)