File size: 3,308 Bytes
dc54faf
 
 
1dbb9f0
c298b5c
 
7cc71c0
4c8a6f3
56abc69
 
 
 
260dc47
dc54faf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
aaf7aaf
 
 
 
dc54faf
 
4c8a6f3
2678939
56abc69
 
 
4c8a6f3
56abc69
9875a43
6e9cf31
 
 
 
 
 
 
 
 
 
4c8a6f3
a0ebadb
c88d72c
 
 
 
a3ad85c
dc54faf
69c15f2
a925a65
49fdcd9
6e9cf31
4848afa
6e9cf31
c88d72c
e694753
 
 
 
 
 
 
 
 
 
 
 
c88d72c
2aa1606
dc54faf
 
56abc69
 
a925a65
 
56abc69
 
dc54faf
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
import gradio as gr
from gpt4all import GPT4All
from huggingface_hub import hf_hub_download
import faiss
#from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_huggingface import HuggingFaceEmbeddings
import numpy as np
from pypdf import PdfReader
from gradio_pdf import PDF
from pdf2image import convert_from_path
from transformers import pipeline
from pathlib import Path
from langchain.vectorstores import Chroma

title = "Mistral-7B-Instruct-GGUF Run On CPU-Basic Free Hardware"

description = """
🔎 [Mistral AI's Mistral 7B Instruct v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1) [GGUF format model](https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-GGUF) , 4-bit quantization balanced quality gguf version, running on CPU. English Only (Also support other languages but the quality's not good). Using [GitHub - llama.cpp](https://github.com/ggerganov/llama.cpp) [GitHub - gpt4all](https://github.com/nomic-ai/gpt4all). 
🔨 Running on CPU-Basic free hardware. Suggest duplicating this space to run without a queue. 
Mistral does not support system prompt symbol (such as ```<<SYS>>```) now, input your system prompt in the first message if you need. Learn more: [Guardrailing Mistral 7B](https://docs.mistral.ai/usage/guardrailing). 
"""

"""
[Model From TheBloke/Mistral-7B-Instruct-v0.1-GGUF](https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-GGUF)
[Mistral-instruct-v0.1 System prompt](https://docs.mistral.ai/usage/guardrailing)
"""

model_path = "models"
model_name = "mistral-7b-instruct-v0.1.Q4_K_M.gguf"

hf_hub_download(repo_id="TheBloke/Mistral-7B-Instruct-v0.1-GGUF", filename=model_name, local_dir=model_path, local_dir_use_symlinks=False)

print("Start the model init process")
model = model = GPT4All(model_name, model_path, allow_download = False, device="cpu")


model.config["promptTemplate"] = "[INST] {0} [/INST]"
model.config["systemPrompt"] = "Tu es un assitant et tu dois répondre en français"
model._is_chat_session_activated = False

max_new_tokens = 2048

model_kwargs = {'device': 'cpu'}
encode_kwargs = {'normalize_embeddings': False}
embeddings = HuggingFaceEmbeddings(
    
    model_kwargs=model_kwargs,
    encode_kwargs=encode_kwargs
)

chunk_size = 2048

# creating a pdf reader object

vectordb = Chroma(
    persist_directory="./resource/chroma/",
    embedding_function=embeddings
)

print("Finish the model init process")

def qa(question: str) -> str:
    

    docs = vectordb.max_marginal_relevance_search(question,k=3)

    

   # prompt = f"""<s>[INST]
    #        Les informations contextuelles sont ci-dessous.
     #       ---------------------
      #      {docs[0].page_content}
       #     ---------------------
        #    [/INST]
         #   Compte tenu des informations contextuelles et non des connaissances préalables, répondez à la requête.  </s>
          #  [INST] Requête: {question} [/INST]
           # Réponse: 
            #    """

    #outputs = model.generate(prompt=prompt, temp=0.5, top_k = 40, top_p = 1, max_tokens = max_new_tokens)
    return docs[0].page_content #"".join(outputs)


demo = gr.Interface(
    qa,
    [gr.Textbox(label="Question")#, PDF(label="Document")
    ],
    gr.Textbox()
)
if __name__ == "__main__":
    demo.queue(max_size=3).launch()