File size: 1,993 Bytes
53e5b63
 
 
 
 
 
 
 
9ff390a
 
53e5b63
 
 
 
 
 
 
 
 
 
9ff390a
53e5b63
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
tags:
- merge
- mergekit
- lazymergekit
- mlabonne/NeuralBeagle14-7B
- mlabonne/NeuralDaredevil-7B
datasets:
- argilla/distilabel-intel-orca-dpo-pairs
---

# DareBeagle-7B

DareBeagle-7B is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [mlabonne/NeuralBeagle14-7B](https://huggingface.co/mlabonne/NeuralBeagle14-7B)
* [mlabonne/NeuralDaredevil-7B](https://huggingface.co/mlabonne/NeuralDaredevil-7B)

As an experiment to find the best base merge to further fine-tuning, expect a lot of experiments named using parts of the component models until a clear winner emerges in the benchmarks

In this case merging the DPO versions of 2 merge models with different characterisics to meassure what capabilities remain or improve

## 🧩 Configuration

```yaml
slices:
  - sources:
      - model: mlabonne/NeuralBeagle14-7B
        layer_range: [0, 32]
      - model: mlabonne/NeuralDaredevil-7B
        layer_range: [0, 32]

merge_method: slerp
base_model: mlabonne/NeuralBeagle14-7B

parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5 # fallback for rest of tensors
tokenizer_source: union

dtype: bfloat16
```

## 💻 Usage

```python
!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "flemmingmiguel/DareBeagle-7B"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
```