magoemu/lora_structeval_t_qwen3_4b_v5_parammatch

lora_structeval_t_qwen3_4b_v5_parammatch

This repository provides a LoRA adapter fine-tuned from Qwen/Qwen3-4B-Instruct-2507 using QLoRA (4-bit, 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: 1024
• Epochs: 2
• Learning rate: 2e-06
• LoRA: r=128, alpha=128, dropout=0.0

Notes

• Loss is applied to assistant response tokens.
• Optional CoT masking is enabled/disabled by CONFIG.

Sources & Terms

Training data: u-10bei/structured_data_with_cot_dataset_512_v4 Please follow the dataset and base model licenses/terms.