qwen3-4b-advanced-sft-v2-merged

This repository provides a merged model fine-tuned from Qwen/Qwen3-4B-Instruct-2507.

• Dataset: u-10bei/sft_alfworld_trajectory_dataset_v5
• Method: LoRA SFT (merged into base model)
• Max sequence length: 2048
• Epochs: 2
• Learning rate: 3e-07
• LoRA: r=96, alpha=192

vLLM Compatibility

• LoRA adapter has been merged
• No tokenizer vocabulary modification
• Intended for AgentBench Advanced evaluation

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "deepkick/qwen3-4b-advanced-sft-v2-merged"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto"
)

Data / License Notes

• Dataset: u-10bei/sft_alfworld_trajectory_dataset_v5 Please refer to the dataset page for license and terms.
• Base model: Qwen/Qwen3-4B-Instruct-2507 Please comply with the base model's original terms of use.