Qwen3-0.6B-wyc
This repository contains a fine-tuned, fully merged model based on
Qwen/Qwen3-0.6B.
The model retains the original Qwen3 architecture and tokenizer. It contains 596,049,920 parameters and can be loaded directly with standard Hugging Face Transformers APIs. No LoRA adapter, custom Python module, external classifier, or remote service is required.
Requirements
python -m pip install -U "transformers>=4.51.0" accelerate torch
hf auth login
This is a gated repository. The Hugging Face account used for downloading must have access to the repository.
Quick start
from transformers import AutoModelForCausalLM, AutoTokenizer
repo_id = "YuenWu/Qwen3-0.6B-wyc"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForCausalLM.from_pretrained(
repo_id,
torch_dtype="auto",
device_map="auto",
)
messages = [
{"role": "user", "content": "Solve: 2 + 3."},
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True,
)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(**inputs, max_new_tokens=256)
new_tokens = generated_ids[:, inputs.input_ids.shape[1]:]
response = tokenizer.batch_decode(new_tokens, skip_special_tokens=True)[0]
print(response)
Set enable_thinking=False in apply_chat_template for non-thinking mode.
Model details
- Base model:
Qwen/Qwen3-0.6B - Architecture:
Qwen3ForCausalLM - Parameter count: 596,049,920
- Weight format: merged BF16
safetensors - Loading method: standard
AutoTokenizerandAutoModelForCausalLM - License: Apache 2.0, following the base model license
Intended use and limitations
This model was produced for research and evaluation of safety alignment, mathematical reasoning, and general instruction following. Model outputs may still be incorrect or unsafe. Applications should independently validate outputs and apply safeguards appropriate to their deployment context.