File size: 1,732 Bytes
fbac662
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5727d2d
fbac662
 
 
 
 
 
 
 
 
5727d2d
fbac662
5727d2d
 
 
fbac662
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5727d2d
fbac662
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
---
base_model: Qwen/Qwen2.5-7B-Instruct
datasets:
- u-10bei/sft_alfworld_trajectory_dataset_v5
- u-10bei/dbbench_sft_dataset_react_v4
language:
- en
license: apache-2.0
library_name: peft
pipeline_tag: text-generation
tags:
- lora
- agent
- tool-use
- alfworld
- dbbench
---

# qwen2.5-7b-alf-dbb-merged-final

This repository provides a **merged full model** based on
**Qwen/Qwen2.5-7B-Instruct**.

## Model Construction Pipeline

1. Train LoRA adapter on ALFWorld
2. Train LoRA adapter on DBBench
3. Merge adapters using `ties` (density=0.1)
4. Apply additional stabilization fine-tuning (LoRA)
5. Merge final adapter into base model

This repository contains **full merged weights (no adapter required)**.

## Final Training Configuration

- Base model: Qwen/Qwen2.5-7B-Instruct
- Merge method: ties
- Merge density: 0.1
- Final stage epochs: 1
- Learning rate: 1e-05
- Final LoRA: r=16, alpha=16
- Max sequence length: 2024


## Datasets

- u-10bei/sft_alfworld_trajectory_dataset_v5
- u-10bei/dbbench_sft_dataset_react_v4

Additional distilled datasets were optionally included.


## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "takayosh/agentbenchfinal"

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

## Sources & Terms (IMPORTANT)

Training data:
- u-10bei/sft_alfworld_trajectory_dataset_v5
- u-10bei/dbbench_sft_dataset_react_v4

Dataset License: MIT License. This dataset is used and distributed under the terms of the MIT License.
Compliance: Users must comply with the MIT license (including copyright notice) and the base model's original terms of use.