wenqiglantz commited on
Commit
b5d5813
1 Parent(s): f2ed7c4

Upload folder using huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +64 -0
README.md ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - merge
5
+ - mergekit
6
+ - lazymergekit
7
+ - stabilityai/stablecode-instruct-alpha-3b
8
+ - stablelm-zephyr-3b
9
+ base_model:
10
+ - stabilityai/stablecode-instruct-alpha-3b
11
+ - stablelm-zephyr-3b
12
+ ---
13
+
14
+ # stable-3b-slerp
15
+
16
+ stable-3b-slerp is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
17
+ * [stabilityai/stablecode-instruct-alpha-3b](https://huggingface.co/stabilityai/stablecode-instruct-alpha-3b)
18
+ * [stablelm-zephyr-3b](https://huggingface.co/stablelm-zephyr-3b)
19
+
20
+ ## 🧩 Configuration
21
+
22
+ ```yaml
23
+ slices:
24
+ - sources:
25
+ - model: stabilityai/stablecode-instruct-alpha-3b
26
+ layer_range: [0, 32]
27
+ - model: stablelm-zephyr-3b
28
+ layer_range: [0, 32]
29
+ merge_method: slerp
30
+ base_model: stabilityai/stablecode-instruct-alpha-3b
31
+ parameters:
32
+ t:
33
+ - filter: self_attn
34
+ value: [0, 0.5, 0.3, 0.7, 1]
35
+ - filter: mlp
36
+ value: [1, 0.5, 0.7, 0.3, 0]
37
+ - value: 0.5
38
+ dtype: bfloat16
39
+ ```
40
+
41
+ ## 💻 Usage
42
+
43
+ ```python
44
+ !pip install -qU transformers accelerate
45
+
46
+ from transformers import AutoTokenizer
47
+ import transformers
48
+ import torch
49
+
50
+ model = "wenqiglantz/stable-3b-slerp"
51
+ messages = [{"role": "user", "content": "What is a large language model?"}]
52
+
53
+ tokenizer = AutoTokenizer.from_pretrained(model)
54
+ prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
55
+ pipeline = transformers.pipeline(
56
+ "text-generation",
57
+ model=model,
58
+ torch_dtype=torch.float16,
59
+ device_map="auto",
60
+ )
61
+
62
+ outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
63
+ print(outputs[0]["generated_text"])
64
+ ```