automerger commited on
Commit
ca198a7
1 Parent(s): fc7af83

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +63 -0
README.md ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - merge
5
+ - mergekit
6
+ - lazymergekit
7
+ base_model:
8
+ - mayacinka/yam-jom-7B-slerp
9
+ - yam-peleg/Experiment28-7B
10
+ ---
11
+
12
+ # YamExperiment28-7B
13
+
14
+ YamExperiment28-7B is an automated merge created by [Maxime Labonne](https://huggingface.co/mlabonne) using the following configuration.
15
+ * [mayacinka/yam-jom-7B-slerp](https://huggingface.co/mayacinka/yam-jom-7B-slerp)
16
+ * [yam-peleg/Experiment28-7B](https://huggingface.co/yam-peleg/Experiment28-7B)
17
+
18
+ ## 🧩 Configuration
19
+
20
+ ```yaml
21
+ slices:
22
+ - sources:
23
+ - model: mayacinka/yam-jom-7B-slerp
24
+ layer_range: [0, 32]
25
+ - model: yam-peleg/Experiment28-7B
26
+ layer_range: [0, 32]
27
+ merge_method: slerp
28
+ base_model: mayacinka/yam-jom-7B-slerp
29
+ parameters:
30
+ t:
31
+ - filter: self_attn
32
+ value: [0, 0.5, 0.3, 0.7, 1]
33
+ - filter: mlp
34
+ value: [1, 0.5, 0.7, 0.3, 0]
35
+ - value: 0.5
36
+ dtype: bfloat16
37
+ random_seed: 0
38
+ ```
39
+
40
+ ## 💻 Usage
41
+
42
+ ```python
43
+ !pip install -qU transformers accelerate
44
+
45
+ from transformers import AutoTokenizer
46
+ import transformers
47
+ import torch
48
+
49
+ model = "automerger/YamExperiment28-7B"
50
+ messages = [{"role": "user", "content": "What is a large language model?"}]
51
+
52
+ tokenizer = AutoTokenizer.from_pretrained(model)
53
+ prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
54
+ pipeline = transformers.pipeline(
55
+ "text-generation",
56
+ model=model,
57
+ torch_dtype=torch.float16,
58
+ device_map="auto",
59
+ )
60
+
61
+ outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
62
+ print(outputs[0]["generated_text"])
63
+ ```