Aznaur/tbench-qwen-sft-multitask-nat-v1

T-Bench Qwen SFT Multi-Task NAT v1

Model Description

This model is fine-tuned from Qwen3-8B using Negative-Aware Training (NAT) on multiple terminal bench tasks.

Training Details

• Base Model: Qwen/Qwen3-8B
• Training Method: Negative-Aware Training (NAT)
• Tasks: 5 tasks (fix-git, log-summary-date-ranges, pypi-server, regex-log, cancel-async-tasks)
• Epochs: 300
• Learning Rate: 5e-5
• Batch Size: 2

Dataset Composition

• Total samples: 26 per epoch
• Positive examples: 20 (4 per task)
• Negative examples: 6 (from fix-git only)

NAT Strategy

Negative examples teach universal anti-patterns:

1. Hallucinated arguments (message_title, message_description)
2. Looping behavior after task completion
3. Wrong command format

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Aznaur/tbench-qwen-sft-multitask-nat-v1")
tokenizer = AutoTokenizer.from_pretrained("Aznaur/tbench-qwen-sft-multitask-nat-v1")

Performance

Trained for 300 epochs with NAT to improve tool usage and avoid common failure patterns.

Paper Reference

Based on "Learning From Failure: Integrating Negative Examples when Fine-tuning Large Language Models as Agents" (arXiv 2402.11651)