yujiepan commited on
Commit
c2e04ae
·
verified ·
1 Parent(s): 0533bfd

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +49 -0
README.md ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pipeline_tag: text-generation
3
+ inference: true
4
+ widget:
5
+ - text: 'Hello!'
6
+ example_title: Hello world
7
+ group: Python
8
+ library_name: transformers
9
+ ---
10
+
11
+ This model is randomly initialized, using the config from [deepseek-ai/deepseek-llm-67b-chat](https://huggingface.co/deepseek-ai/deepseek-llm-67b-chat) but with smaller size.
12
+ Note the model is in float16.
13
+
14
+ Codes:
15
+ ```python
16
+ from transformers import pipeline
17
+ from huggingface_hub import create_repo, upload_folder
18
+ import torch
19
+ import transformers
20
+ import os
21
+
22
+ model_id = 'mistralai/Mixtral-8x7B-Instruct-v0.1'
23
+ save_path = '/tmp/yujiepan/mixtral-tiny-random'
24
+ repo_id = 'yujiepan/mixtral-tiny-random'
25
+
26
+ config = transformers.AutoConfig.from_pretrained(model_id)
27
+ config.hidden_size = 4
28
+ config.intermediate_size = 8
29
+ config.num_attention_heads = 4
30
+ config.num_experts_per_tok = 2
31
+ config.num_hidden_layers = 2
32
+ config.num_key_value_heads = 2
33
+ config.num_local_experts = 8
34
+ print(config)
35
+
36
+ tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
37
+ tokenizer.save_pretrained(save_path)
38
+
39
+ model = transformers.AutoModelForCausalLM.from_config(config, torch_dtype=torch.float16)
40
+ model = model.half()
41
+
42
+ pipe = pipeline('text-generation', model=model, tokenizer=tokenizer, do_sample=False, device='cuda')
43
+ print(pipe('Hello World!'))
44
+
45
+ model.save_pretrained(save_path)
46
+
47
+ create_repo(repo_id, exist_ok=True)
48
+ upload_folder(repo_id=repo_id, folder_path=save_path)
49
+ ```