---
base_model: Qwen/Qwen2.5-7B-Instruct
datasets: []
language:
- en
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
tags:
- agentbench
- merged-model
---

# qwen2.5-7b-instruct-sft-v1

This repository provides a merged full model produced by supervised fine-tuning for task-oriented instruction following.

## Training Objective

Improve instruction following, action consistency, and response reliability in practical workflows.

## Training Configuration

- Method: SFT (TRL SFTTrainer + Transformers, full-model)
- Max sequence length: 1024
- Max steps: 6
- Epochs: 1
- Learning rate: 2e-6
- Per-device train batch size: 1
- Gradient accumulation steps: 8
- Effective global batch size: 8
- Training schedule: Step-based (`max_steps` takes precedence in Trainer).

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "uchkw/qwen2.5-7b-instruct-sft-v1"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
```

## Training Data / Sources & License (IMPORTANT)

- Datasets: rule-based original synthetic data
- Compliance: Users must comply with the base model's terms of use.

## Training Plan Summary

- Objective: Push DB score upward aggressively while preserving ALF above baseline.
- Strategy:
    - Increase DB rows strongly and keep action rows moderate.
    - Use stronger SQL stage and very light action stage.