qwen3-0.6b-llmft-flavor

llmft/qwen3-0.6b-llmft-flavor

LLMFT flavor of Qwen3-0.6B for free on-device Android chat.

This repository contains a LoRA adapter trained with llmfinetuning on top of Qwen/Qwen3-0.6B.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = "Qwen/Qwen3-0.6B"
adapter = "llmft/qwen3-0.6b-llmft-flavor"

tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype="auto", device_map="auto")
model = PeftModel.from_pretrained(model, adapter)

messages = [{"role": "user", "content": "What is LoRA?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
out = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(out[0], skip_special_tokens=True))

Training

• Framework: llmft (PEFT LoRA + TRL SFT)
• Mode: json
• Base model: Qwen/Qwen3-0.6B

Notes

• This is an adapter, not a full foundation model like Qwen/Qwen3-0.6B.
• Users must download the base model + this adapter.
• For a single merged checkpoint, re-run push with --merge.