bofenghuang
commited on
Commit
•
615cfe6
1
Parent(s):
3a8dd65
Add doc
Browse files- README.md +144 -0
- logo_v2.jpg +0 -0
README.md
ADDED
@@ -0,0 +1,144 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: llama2
|
3 |
+
language: fr
|
4 |
+
pipeline_tag: text-generation
|
5 |
+
inference: false
|
6 |
+
tags:
|
7 |
+
- LLM
|
8 |
+
- llama-2
|
9 |
+
- finetuned
|
10 |
+
---
|
11 |
+
|
12 |
+
<p align="center" width="100%">
|
13 |
+
<img src="https://huggingface.co/bofenghuang/vigogne-2-70b-chat/resolve/main/logo_v2.jpg" alt="Vigogne" style="width: 30%; min-width: 300px; display: block; margin: auto;">
|
14 |
+
</p>
|
15 |
+
|
16 |
+
# Vigogne-2-70B-Chat: A Llama-2-based French Chat LLM
|
17 |
+
|
18 |
+
Vigogne-2-70B-Chat is a French chat LLM, based on [Llama-2-70B](https://ai.meta.com/llama), optimized to generate helpful and coherent responses in conversations with users.
|
19 |
+
|
20 |
+
Check out our [release blog](https://github.com/bofenghuang/vigogne/blob/main/blogs/2023-08-17-vigogne-chat-v2_0.md) and [GitHub repository](https://github.com/bofenghuang/vigogne) for more information.
|
21 |
+
|
22 |
+
**Usage and License Notices**: Vigogne-2-70B-Chat follows Llama-2's [usage policy](https://ai.meta.com/llama/use-policy). A significant portion of the training data is distilled from GPT-3.5-Turbo and GPT-4, kindly use it cautiously to avoid any violations of OpenAI's [terms of use](https://openai.com/policies/terms-of-use).
|
23 |
+
|
24 |
+
## Prompt Template
|
25 |
+
|
26 |
+
We used a prompt template adapted from the chat format of Llama-2.
|
27 |
+
|
28 |
+
You can apply this formatting using the [chat template](https://huggingface.co/docs/transformers/main/chat_templating) through the `apply_chat_template()` method.
|
29 |
+
|
30 |
+
```python
|
31 |
+
from transformers import AutoTokenizer
|
32 |
+
|
33 |
+
tokenizer = AutoTokenizer.from_pretrained("bofenghuang/vigogne-2-70b-chat")
|
34 |
+
|
35 |
+
conversation = [
|
36 |
+
{"role": "user", "content": "Bonjour ! Comment ça va aujourd'hui ?"},
|
37 |
+
{"role": "assistant", "content": "Bonjour ! Je suis une IA, donc je n'ai pas de sentiments, mais je suis prêt à vous aider. Comment puis-je vous assister aujourd'hui ?"},
|
38 |
+
{"role": "user", "content": "Quelle est la hauteur de la Tour Eiffel ?"},
|
39 |
+
{"role": "assistant", "content": "La Tour Eiffel mesure environ 330 mètres de hauteur."},
|
40 |
+
{"role": "user", "content": "Comment monter en haut ?"},
|
41 |
+
]
|
42 |
+
|
43 |
+
print(tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True))
|
44 |
+
```
|
45 |
+
|
46 |
+
You will get
|
47 |
+
|
48 |
+
```
|
49 |
+
<s>[INST] <<SYS>>
|
50 |
+
Vous êtes Vigogne, un assistant IA créé par Zaion Lab. Vous suivez extrêmement bien les instructions. Aidez autant que vous le pouvez.
|
51 |
+
<</SYS>>
|
52 |
+
|
53 |
+
Bonjour ! Comment ça va aujourd'hui ? [/INST] Bonjour ! Je suis une IA, donc je n'ai pas de sentiments, mais je suis prêt à vous aider. Comment puis-je vous assister aujourd'hui ? </s>[INST] Quelle est la hauteur de la Tour Eiffel ? [/INST] La Tour Eiffel mesure environ 330 mètres de hauteur. </s>[INST] Comment monter en haut ? [/INST]
|
54 |
+
```
|
55 |
+
|
56 |
+
## Usage
|
57 |
+
|
58 |
+
<!-- ### Inference using the quantized versions
|
59 |
+
|
60 |
+
The quantized versions of this model are generously provided by [TheBloke](https://huggingface.co/TheBloke)!
|
61 |
+
|
62 |
+
- AWQ for GPU inference: [TheBloke/Vigogne-2-70B-Chat-AWQ](https://huggingface.co/TheBloke/Vigogne-2-70B-Chat-AWQ)
|
63 |
+
- GTPQ for GPU inference: [TheBloke/Vigogne-2-70B-Chat-GPTQ](https://huggingface.co/TheBloke/Vigogne-2-70B-Chat-GPTQ)
|
64 |
+
- GGUF for CPU+GPU inference: [TheBloke/Vigogne-2-70B-Chat-GGUF](https://huggingface.co/TheBloke/Vigogne-2-70B-Chat-GGUF)
|
65 |
+
|
66 |
+
These versions facilitate testing and development with various popular frameworks, including [AutoAWQ](https://github.com/casper-hansen/AutoAWQ), [vLLM](https://github.com/vllm-project/vllm), [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ), [GPTQ-for-LLaMa](https://github.com/qwopqwop200/GPTQ-for-LLaMa), [llama.cpp](https://github.com/ggerganov/llama.cpp), [text-generation-webui](https://github.com/oobabooga/text-generation-webui), and more. -->
|
67 |
+
|
68 |
+
### Inference using the unquantized model with 🤗 Transformers
|
69 |
+
|
70 |
+
```python
|
71 |
+
from typing import Dict, List, Optional
|
72 |
+
import torch
|
73 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig, TextStreamer
|
74 |
+
|
75 |
+
model_name_or_path = "bofenghuang/vigogne-2-70b-chat"
|
76 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, padding_side="right", use_fast=False)
|
77 |
+
model = AutoModelForCausalLM.from_pretrained(model_name_or_path, torch_dtype=torch.float16, device_map="auto")
|
78 |
+
|
79 |
+
streamer = TextStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
80 |
+
|
81 |
+
|
82 |
+
def chat(
|
83 |
+
query: str,
|
84 |
+
history: Optional[List[Dict]] = None,
|
85 |
+
temperature: float = 0.7,
|
86 |
+
top_p: float = 1.0,
|
87 |
+
top_k: float = 0,
|
88 |
+
repetition_penalty: float = 1.1,
|
89 |
+
max_new_tokens: int = 1024,
|
90 |
+
**kwargs,
|
91 |
+
):
|
92 |
+
if history is None:
|
93 |
+
history = []
|
94 |
+
|
95 |
+
history.append({"role": "user", "content": query})
|
96 |
+
|
97 |
+
input_ids = tokenizer.apply_chat_template(history, return_tensors="pt").to(model.device)
|
98 |
+
input_length = input_ids.shape[1]
|
99 |
+
|
100 |
+
generated_outputs = model.generate(
|
101 |
+
input_ids=input_ids,
|
102 |
+
generation_config=GenerationConfig(
|
103 |
+
temperature=temperature,
|
104 |
+
do_sample=temperature > 0.0,
|
105 |
+
top_p=top_p,
|
106 |
+
top_k=top_k,
|
107 |
+
repetition_penalty=repetition_penalty,
|
108 |
+
max_new_tokens=max_new_tokens,
|
109 |
+
pad_token_id=tokenizer.eos_token_id,
|
110 |
+
**kwargs,
|
111 |
+
),
|
112 |
+
streamer=streamer,
|
113 |
+
return_dict_in_generate=True,
|
114 |
+
)
|
115 |
+
|
116 |
+
generated_tokens = generated_outputs.sequences[0, input_length:]
|
117 |
+
generated_text = tokenizer.decode(generated_tokens, skip_special_tokens=True)
|
118 |
+
|
119 |
+
history.append({"role": "assistant", "content": generated_text})
|
120 |
+
|
121 |
+
return generated_text, history
|
122 |
+
|
123 |
+
|
124 |
+
# 1st round
|
125 |
+
response, history = chat("Un escargot parcourt 100 mètres en 5 heures. Quelle est sa vitesse ?", history=None)
|
126 |
+
|
127 |
+
# 2nd round
|
128 |
+
response, history = chat("Quand il peut dépasser le lapin ?", history=history)
|
129 |
+
|
130 |
+
# 3rd round
|
131 |
+
response, history = chat("Écris une histoire imaginative qui met en scène une compétition de course entre un escargot et un lapin.", history=history)
|
132 |
+
```
|
133 |
+
|
134 |
+
You can also use the Google Colab Notebook provided below.
|
135 |
+
|
136 |
+
<a href="https://colab.research.google.com/github/bofenghuang/vigogne/blob/main/notebooks/infer_chat.ipynb" target="_blank"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a>
|
137 |
+
|
138 |
+
## Limitations
|
139 |
+
|
140 |
+
Vigogne is still under development, and there are many limitations that have to be addressed. Please note that it is possible that the model generates harmful or biased content, incorrect information or generally unhelpful answers.
|
141 |
+
|
142 |
+
## Acknowledgements
|
143 |
+
|
144 |
+
The model training was conducted on the [Jean-Zay supercomputer](http://www.idris.fr/eng/jean-zay/jean-zay-presentation-eng.html) at GENCI, and we extend our gratitude to the IDRIS team for their responsive support throughout the project.
|
logo_v2.jpg
ADDED