Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
tags:
|
3 |
+
- merge
|
4 |
+
- mergekit
|
5 |
+
- lazymergekit
|
6 |
+
- abideen/AlphaMonarch-dora
|
7 |
+
base_model:
|
8 |
+
- abideen/AlphaMonarch-dora
|
9 |
+
license: cc-by-nc-4.0
|
10 |
+
---
|
11 |
+
|
12 |
+
# Spaetzle-v69-7b
|
13 |
+
This is a progressive (mostly dare-ties, but also slerp) merge with the intention of a suitable compromise for English and German local tasks.
|
14 |
+
|
15 |
+
There is also an [unquantized](https://huggingface.co/cstr/Spaetzle-v69-7b) version.
|
16 |
+
|
17 |
+
It achieves (running quantized) in
|
18 |
+
- German EQ Bench: Score (v2_de): 62.59 (Parseable: 171.0).
|
19 |
+
- English EQ Bench: Score (v2): 76.43 (Parseable: 171.0).
|
20 |
+
|
21 |
+
It should work sufficiently well with ChatML prompt template (for all merged models should have seen ChatML prompts at least in DPO stage).
|
22 |
+
|
23 |
+
Spaetzle-v69-7b is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
|
24 |
+
* [abideen/AlphaMonarch-dora](https://huggingface.co/abideen/AlphaMonarch-dora)
|
25 |
+
* [cstr/Spaetzle-v68-7b](https://huggingface.co/cstr/Spaetzle-v68-7b)
|
26 |
+
|
27 |
+
The merge tree in total involves to following original models:
|
28 |
+
- [abideen/AlphaMonarch-dora](https://huggingface.co/abideen/AlphaMonarch-dora)
|
29 |
+
- [mayflowergmbh/Wiedervereinigung-7b-dpo](https://huggingface.co/mayflowergmbh/Wiedervereinigung-7b-dpo)
|
30 |
+
- [flemmingmiguel/NeuDist-Ro-7B](https://huggingface.co/flemmingmiguel/NeuDist-Ro-7B)
|
31 |
+
- [ResplendentAI/Flora_DPO_7B](https://huggingface.co/ResplendentAI/Flora_DPO_7B)
|
32 |
+
- [yleo/EmertonMonarch-7B](https://huggingface.co/yleo/EmertonMonarch-7B)
|
33 |
+
- [occiglot/occiglot-7b-de-en-instruct](https://huggingface.co/occiglot/occiglot-7b-de-en-instruct)
|
34 |
+
- [OpenPipe/mistral-ft-optimized-1227](https://huggingface.co/OpenPipe/mistral-ft-optimized-1227)
|
35 |
+
- [yleo/EmertonMonarch-7B](https://huggingface.co/yleo/EmertonMonarch-7B)
|
36 |
+
- [DiscoResearch/DiscoLM_German_7b_v1](https://huggingface.co/DiscoResearch/DiscoLM_German_7b_v1)
|
37 |
+
- [LeoLM/leo-mistral-hessianai-7b](https://huggingface.co/LeoLM/leo-mistral-hessianai-7b)
|
38 |
+
- [DRXD1000/Phoenix](https://huggingface.co/DRXD1000/Phoenix)
|
39 |
+
- [VAGOsolutions/SauerkrautLM-7b-v1-mistral](https://huggingface.co/VAGOsolutions/SauerkrautLM-7b-v1-mistral)
|
40 |
+
- [malteos/hermeo-7b](https://huggingface.co/malteos/hermeo-7b)
|
41 |
+
- [FelixChao/WestSeverus-7B-DPO-v2](https://huggingface.co/FelixChao/WestSeverus-7B-DPO-v2)
|
42 |
+
- [cognitivecomputations/openchat-3.5-0106-laser](https://huggingface.co/cognitivecomputations/openchat-3.5-0106-laser)
|
43 |
+
|
44 |
+
|
45 |
+
## 🧩 Configuration
|
46 |
+
|
47 |
+
```yaml
|
48 |
+
models:
|
49 |
+
- model: cstr/Spaetzle-v68-7b
|
50 |
+
# no parameters necessary for base model
|
51 |
+
- model: abideen/AlphaMonarch-dora
|
52 |
+
parameters:
|
53 |
+
density: 0.60
|
54 |
+
weight: 0.30
|
55 |
+
merge_method: dare_ties
|
56 |
+
base_model: cstr/Spaetzle-v68-7b
|
57 |
+
parameters:
|
58 |
+
int8_mask: true
|
59 |
+
dtype: bfloat16
|
60 |
+
random_seed: 0
|
61 |
+
tokenizer_source: base
|
62 |
+
```
|
63 |
+
|
64 |
+
## 💻 Usage
|
65 |
+
|
66 |
+
```python
|
67 |
+
!pip install -qU transformers accelerate
|
68 |
+
|
69 |
+
from transformers import AutoTokenizer
|
70 |
+
import transformers
|
71 |
+
import torch
|
72 |
+
|
73 |
+
model = "cstr/Spaetzle-v69-7b"
|
74 |
+
messages = [{"role": "user", "content": "What is a large language model?"}]
|
75 |
+
|
76 |
+
tokenizer = AutoTokenizer.from_pretrained(model)
|
77 |
+
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
78 |
+
pipeline = transformers.pipeline(
|
79 |
+
"text-generation",
|
80 |
+
model=model,
|
81 |
+
torch_dtype=torch.float16,
|
82 |
+
device_map="auto",
|
83 |
+
)
|
84 |
+
|
85 |
+
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
|
86 |
+
print(outputs[0]["generated_text"])
|
87 |
+
```
|