Satyam-Singh commited on
Commit
d215d95
1 Parent(s): 1a25f83

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +124 -1
README.md CHANGED
@@ -1,3 +1,126 @@
1
  ---
2
- license: gpl-3.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: apache-2.0
3
+ language:
4
+ - fr
5
+ - it
6
+ - de
7
+ - es
8
+ - en
9
+ - hi
10
+ - pa
11
+ - ta
12
+ - te
13
+ - mr
14
+ - be
15
+ - ba
16
+ - bn
17
+ - ru
18
+ - ee
19
+ - am
20
+ - ar
21
+ - qu
22
+ - ja
23
+ - ch
24
+ - kw
25
  ---
26
+ # Model Card for LLaVa-8x7B
27
+ The LLaVa-8x7B Large Language Model (LLM) is a pretrained generative Sparse Mixture of Experts. The LLaVa-8x7B outperforms Llama 2 70B on most benchmarks we tested.
28
+
29
+
30
+ ## Warning
31
+ This repo contains weights that are compatible with [vLLM](https://github.com/vllm-project/vllm) serving of the model as well as Hugging Face [transformers](https://github.com/huggingface/transformers) library. It is based on the original LLaVa [torrent release](magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&dn=mixtral-8x7b-32kseqlen&tr=udp%3A%2F%http://2Fopentracker.i2p.rocks%3A6969%2Fannounce&tr=http%3A%2F%http://2Ftracker.openbittorrent.com%3A80%2Fannounce), but the file format and parameter names are different. Please note that model cannot (yet) be instantiated with HF.
32
+
33
+ ## Run the model
34
+
35
+
36
+ ```python
37
+ from transformers import AutoModelForCausalLM, AutoTokenizer
38
+
39
+ model_id = "Satyam-Singh/LLava-8x7B-v0.1"
40
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
41
+
42
+ model = AutoModelForCausalLM.from_pretrained(model_id)
43
+
44
+ text = "Hello my name is"
45
+ inputs = tokenizer(text, return_tensors="pt")
46
+
47
+ outputs = model.generate(**inputs, max_new_tokens=20)
48
+ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
49
+ ```
50
+
51
+ By default, transformers will load the model in full precision. Therefore you might be interested to further reduce down the memory requirements to run the model through the optimizations we offer in HF ecosystem:
52
+
53
+ ### In half-precision
54
+
55
+ Note `float16` precision only works on GPU devices
56
+
57
+ <details>
58
+ <summary> Click to expand </summary>
59
+
60
+ ```diff
61
+ + import torch
62
+ from transformers import AutoModelForCausalLM, AutoTokenizer
63
+
64
+ model_id = "Satyam-Singh/LLava-8x7B-v0.1"
65
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
66
+
67
+ + model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16).to(0)
68
+
69
+ text = "Hello my name is"
70
+ + inputs = tokenizer(text, return_tensors="pt").to(0)
71
+
72
+ outputs = model.generate(**inputs, max_new_tokens=20)
73
+ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
74
+ ```
75
+ </details>
76
+
77
+ ### Lower precision using (8-bit & 4-bit) using `bitsandbytes`
78
+
79
+ <details>
80
+ <summary> Click to expand </summary>
81
+
82
+ ```diff
83
+ + import torch
84
+ from transformers import AutoModelForCausalLM, AutoTokenizer
85
+
86
+ model_id = "Satyam-Singh/LLava-8x7B-v0.1"
87
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
88
+
89
+ + model = AutoModelForCausalLM.from_pretrained(model_id, load_in_4bit=True)
90
+
91
+ text = "Hello my name is"
92
+ + inputs = tokenizer(text, return_tensors="pt").to(0)
93
+
94
+ outputs = model.generate(**inputs, max_new_tokens=20)
95
+ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
96
+ ```
97
+ </details>
98
+
99
+ ### Load the model with Flash Attention 2
100
+
101
+ <details>
102
+ <summary> Click to expand </summary>
103
+
104
+ ```diff
105
+ + import torch
106
+ from transformers import AutoModelForCausalLM, AutoTokenizer
107
+
108
+ model_id = "Satyam-Singh/LLava-8x7B-v0.1"
109
+ tokenizer = AutoTokenizer.from_pretrained(model_id)
110
+
111
+ + model = AutoModelForCausalLM.from_pretrained(model_id, use_flash_attention_2=True)
112
+
113
+ text = "Hello my name is"
114
+ + inputs = tokenizer(text, return_tensors="pt").to(0)
115
+
116
+ outputs = model.generate(**inputs, max_new_tokens=20)
117
+ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
118
+ ```
119
+ </details>
120
+
121
+ ## Notice
122
+ LLava-8x7B-v0.1 is a pretrained base model and therefore does not have any moderation mechanisms.
123
+
124
+
125
+
126
+ # Love From The LLaVa AI & UniVerse Unique AI Team