metadata
license: apache-2.0
base_model: Qwen/Qwen3-4B-Instruct-2507
tags:
- unsloth
- trl
- sft
- qwen
- json-generation
- struct-eval
language:
- ja
- en
qwen3-4b-struct-eval-test
Model Description
Fine-tuned version of Qwen/Qwen3-4B-Instruct-2507 using Unsloth.
Fine-tuning focus: Structured data generation (JSON output)
Training Details
- Base Model: Qwen3-4B-Instruct-2507
- Method: LoRA (rank=16, alpha=32)
- Framework: Unsloth + TRL
- Optimization: 4-bit quantization
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "yuk1chan/qwen3-4b-struct-eval-test"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
prompt = """JSON only:
Name: 田中太郎, Age: 30, Job: エンジニア
{"""
inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.1)
result = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
print(result)
Model Card Author
- yuk1chan
License
Apache 2.0 (inherited from base model)