OsakanaTeishoku/Qwen3-4B-Thinking-2507-reasoning-ja-20260329

概要

DataPilot/Knowledge-QA-SingleTurn-DatasetでSFTし，日本語の入力に対し日本語で思考するようにしたモデルです．コンテキスト長は16384です．

使い方

from transformers import AutoTokenizer, AutoModelForCausalLM, TextStreamer

tokenizer = AutoTokenizer.from_pretrained("OsakanaTeishoku/Qwen3-4B-Thinking-2507-reasoning-ja-20260329")
model = AutoModelForCausalLM.from_pretrained("OsakanaTeishoku/Qwen3-4B-Thinking-2507-reasoning-ja-20260329", dtype="auto", device_map="auto")
messages = [
    {"role": "user", "content": "肉じゃがの作り方を教えて"},
]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

outputs = model.generate(
    **inputs, 
    max_new_tokens=5000, 
    do_sample=True,
    temperature=0.7,
    top_p=0.8,
    top_k=20,
    streamer=streamer,              
)

Uploaded finetuned model

• Developed by: OsakanaTeishoku
• License: apache-2.0
• Finetuned from model : unsloth/Qwen3-4B-Thinking-2507

This qwen3 model was trained 2x faster with Unsloth and Huggingface's TRL library.