structeval-qwen3-4b-sft-try51

＜【課題】ここは自分で記入して下さい＞

This repository provides a LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using LoRA + Unsloth.

This repository contains LoRA adapter weights only. The base model must be loaded separately.

Training Configuration

• Base model: Qwen/Qwen3-4B-Instruct-2507
• Max sequence length: 2048
• Epochs: 2
• Learning rate: 2e-06
• LoRA: r=64, alpha=128

Sources & Terms (IMPORTANT)

Training data (YAML): u-10bei/sft_alfworld_trajectory_dataset_v5 Training data (actually used by training code): /content/alf_try_balanced.jsonl

Note: HF YAML metadata does not allow local file paths in datasets:. If you trained from a locally derived JSONL (e.g., /content/alf_try_balanced.jsonl), it should be described here (as above), not in YAML.