File size: 4,655 Bytes
333db1f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
import os
import gradio as gr
from langchain_huggingface import HuggingFaceEndpoint,HuggingFaceEmbeddings,ChatHuggingFace
from langchain_core.load import dumpd, dumps, load, loads
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.callbacks import StreamingStdOutCallbackHandler

from langchain_chroma import Chroma
from langchain_core.documents import Document  
from langchain_text_splitters import CharacterTextSplitter
from pypdf import PdfReader
import random

token=""
#repo_id = "mistralai/Mistral-7B-Instruct-v0.3"
repo_id = "mistralai/Mixtral-8x7B-Instruct-v0.1"
emb = "sentence-transformers/all-mpnet-base-v2"
hf = HuggingFaceEmbeddings(model_name=emb)
db = Chroma(persist_directory="./chroma_langchain_db")
db.persist()
# Load the document, split it into chunks, embed each chunk and load it into the vector store.
#raw_documents = TextLoader('state_of_the_union.txt').load()
def embed_fn(inp):
    print("Try Embeddings")
    print(inp)
    print("End Embeddings")
    #for eaa in inp:
    text_splitter = CharacterTextSplitter(chunk_size=200, chunk_overlap=10)
    #documents = text_splitter.split_documents([eaa])
    documents = text_splitter.split_text(inp)
    print("documents")
    print(documents)
    print("end documents")
    out_emb= hf.embed_documents(documents)
    #chain = history[:-1]
    string_representation = dumps(out_emb, pretty=True)
    print(string_representation)
    #db = Chroma(collection_name="test1", embedding_function=HuggingFaceEmbeddings())
    db.from_texts(documents,HuggingFaceEmbeddings(model_name=emb))
    #from_documents(documents, HuggingFaceEmbeddings)
    print("DB")
    print(db)
    print("end DB")
    #return db
def proc_doc(doc_in):
    for doc in doc_in:
        if doc.endswith(".txt"):
            yield [["",f"Loading Document: {doc}"]]
            outp = read_txt(doc)
            embed_fn(outp)
            yield [["","Loaded"]]
        elif doc.endswith(".pdf"):
            yield [["",f"Loading Document: {doc}"]]
            outp = read_pdf(doc)
            embed_fn(outp)
            yield [["","Loaded"]]


def read_txt(txt_path):
    text=""
    with open(txt_path,"r") as f:
        text = f.read()
    f.close()
    return text

def read_pdf(pdf_path):
    text=""
    reader = PdfReader(f'{pdf_path}')
    number_of_pages = len(reader.pages)
    for i in range(number_of_pages):
        page = reader.pages[i]
        text = f'{text}\n{page.extract_text()}'
    return text
def run_llm(input_text,history):
    MAX_TOKENS=20000
    qur= hf.embed_query(input_text)
    docs = db.similarity_search_by_vector(qur, k=3)
    
    '''if len(docs) >2:

    doc_list = str(docs).split(" ")
    if len(doc_list) > MAX_TOKENS:
        doc_cnt = int(len(doc_list) / MAX_TOKENS)
        print(doc_cnt)
        for ea in doc_cnt:'''


    print(docs)

    callbacks = [StreamingStdOutCallbackHandler()]  
    llm = HuggingFaceEndpoint(  
    endpoint_url=repo_id,  
    max_new_tokens=2056,  
    seed=random.randint(1,99999999999),
    top_k=10,  
    top_p=0.95,  
    typical_p=0.95,  
    temperature=0.01,  
    repetition_penalty=1.03,  
    #callbacks=callbacks,  
    streaming=True,  
    huggingfacehub_api_token=token,
    )  


    '''llm=HuggingFaceEndpoint(
        endpoint_url=repo_id,
        streaming=True,
        max_new_tokens=2400,
        huggingfacehub_api_token=token)'''
    print(input_text)
    print(history)
    out=""
    #prompt = ChatPromptTemplate.from_messages(
    sys_prompt = f"Use this data to help answer users questions: {str(docs)}"
    user_prompt = f"{input_text}"
    prompt=[
        {"role": "system", "content": f"[INST] Use this data to help answer users questions: {str(docs)} [/INST]"},
        {"role": "user", "content": f"[INST]{input_text}[/INST]"},
    ]
    #chat = ChatHuggingFace(llm=llm, verbose=True)
    messages = [  
    ("system", f"[INST] Use this data to help answer users questions: {str(docs)} [/INST]"),  
    ("user", f"[INST]{input_text}[/INST]"),  
    ]  
    
    #yield(llm.invoke(prompt))
    
    t=llm.invoke(prompt)
    for chunk in t:
        out+=chunk
        yield out
      
    
css="""
#component-0 {
    height:400px;
}
"""

with gr.Blocks(css=css) as app:
    data=gr.State()
    with gr.Column():
        #input_text = gr.Textbox(label="You: ")
        chat = gr.ChatInterface(
            fn=run_llm,
            type="tuples",
            concurrency_limit=20,
            
        )
    with gr.Row():
        msg=gr.HTML()
        file_in=gr.Files(file_count="multiple")
    file_in.change(proc_doc, file_in, msg)
        #btn = gr.Button("Generate")
app.queue().launch()