Spaces:
Sleeping
Sleeping
init
Browse files- app.py +132 -0
- requirements.txt +155 -0
app.py
ADDED
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Imports
|
2 |
+
from langchain.document_loaders import PyPDFLoader
|
3 |
+
import os
|
4 |
+
from langchain.chains import RetrievalQA, ConversationalRetrievalChain
|
5 |
+
from langchain.indexes import VectorstoreIndexCreator
|
6 |
+
from langchain.text_splitter import CharacterTextSplitter, TokenTextSplitter
|
7 |
+
from langchain.embeddings import OpenAIEmbeddings, HuggingFaceEmbeddings
|
8 |
+
from langchain.vectorstores import Chroma
|
9 |
+
from langchain import HuggingFacePipeline
|
10 |
+
from langchain.chat_models import ChatOpenAI
|
11 |
+
from dotenv import load_dotenv
|
12 |
+
from langchain.memory import ConversationBufferMemory, ConversationTokenBufferMemory
|
13 |
+
import gradio as gr
|
14 |
+
|
15 |
+
# Funcion de carga de la api key
|
16 |
+
def process_key(api_key):
|
17 |
+
os.environ['OPENAI_API_KEY'] = api_key
|
18 |
+
|
19 |
+
def load_pdf(file):
|
20 |
+
name_file = file.name
|
21 |
+
print(file.name)
|
22 |
+
loader = PyPDFLoader(file.name)
|
23 |
+
documents = loader.load()
|
24 |
+
print(documents)
|
25 |
+
# Creo el objeto que permite dividir el texto en chunks
|
26 |
+
text_splitter = CharacterTextSplitter(chunk_size=1024, chunk_overlap=64)
|
27 |
+
# Esto lo que hace es dividir el texto en chunks de 2048 caracteres con un overlap de 128 caracteres
|
28 |
+
texts = text_splitter.split_documents(documents)
|
29 |
+
# Genero el objeto que crea los embeddings
|
30 |
+
# Nota: Estos embeddings son gratuitos a diferencia de los de OpenAI
|
31 |
+
embeddings = HuggingFaceEmbeddings()
|
32 |
+
# Defino el modelo de lenguaje
|
33 |
+
llm = ChatOpenAI(model='gpt-3.5-turbo', temperature=0.0, max_tokens=1000)
|
34 |
+
# Creo la base de datos de vectores
|
35 |
+
global vectorstore
|
36 |
+
vectorstore = Chroma.from_documents(texts, embeddings)
|
37 |
+
# Defino la memoria
|
38 |
+
|
39 |
+
global memory
|
40 |
+
# La definicion de Memoria no es trivial, es bastante compleja de hecho se deben especificar bien todos los parameteros para que no de error
|
41 |
+
memory = ConversationTokenBufferMemory(llm=llm,
|
42 |
+
memory_key="chat_history",
|
43 |
+
input_key='question',
|
44 |
+
output_key='answer',
|
45 |
+
max_token_limit=1000,
|
46 |
+
return_messages=False)
|
47 |
+
# Defino la cadena de qa
|
48 |
+
global qa
|
49 |
+
qa = ConversationalRetrievalChain.from_llm(llm,
|
50 |
+
vectorstore.as_retriever(search_kwargs={'k': 3}), # Este parametro especifica cuantos chunks se van a recuperar
|
51 |
+
return_source_documents=True,
|
52 |
+
verbose=True,
|
53 |
+
chain_type='stuff',
|
54 |
+
memory=memory,
|
55 |
+
max_tokens_limit=2500,
|
56 |
+
get_chat_history=lambda h: h)
|
57 |
+
return 'Done'
|
58 |
+
|
59 |
+
# Funcion que ejecuta LLM y responde la pregunta
|
60 |
+
def answer_question(question):
|
61 |
+
result = qa(inputs={'question': question})
|
62 |
+
pages = [x.metadata['page'] for i, x in enumerate(result['source_documents'])]
|
63 |
+
return result['answer'], pages
|
64 |
+
|
65 |
+
# Funcion que pega las respuestas anteriores en el objeto Chat bot
|
66 |
+
def bot(history):
|
67 |
+
res = qa(
|
68 |
+
{
|
69 |
+
'question': history[-1][0],
|
70 |
+
'chat_history': history[:-1]
|
71 |
+
}
|
72 |
+
)
|
73 |
+
history[-1][1] = res['answer']
|
74 |
+
return history
|
75 |
+
|
76 |
+
# Agrego el texto a la historia del chat
|
77 |
+
def add_text(history, text):
|
78 |
+
history = history + [(text, None)]
|
79 |
+
return history, ""
|
80 |
+
|
81 |
+
# Analizar como parsea las ecuaciones
|
82 |
+
with gr.Blocks() as demo:
|
83 |
+
with gr.Tab(label='Load PDF'):
|
84 |
+
with gr.Row():
|
85 |
+
with gr.Column():
|
86 |
+
open_ai_key = gr.Textbox(label='Ingresa tu api key de Open AI', type='password')
|
87 |
+
with gr.Row():
|
88 |
+
with gr.Column(scale=0.4):
|
89 |
+
api_key_button = gr.Button('Enviar', variant='primary')
|
90 |
+
with gr.Row():
|
91 |
+
pdf_file = gr.File(label='PDF file')
|
92 |
+
# Esta linea esta para probar si el calculo se realiza
|
93 |
+
emb = gr.Textbox(label='Calculo de Embeddings, por favor espere...')
|
94 |
+
# send_pdf = gr.Button(label='Load PDF').style(full_width=False)
|
95 |
+
with gr.Row():
|
96 |
+
with gr.Column(scale=0.50):
|
97 |
+
send_pdf = gr.Button(label='Load PDF')
|
98 |
+
send_pdf.click(load_pdf, pdf_file, emb)
|
99 |
+
with gr.Tab(label='Galicia QA Demo'):
|
100 |
+
chatbot = gr.Chatbot([],
|
101 |
+
elem_id="chatbot",
|
102 |
+
label='Document GPT').style(height=500)
|
103 |
+
with gr.Row():
|
104 |
+
with gr.Column(scale=0.80):
|
105 |
+
txt = gr.Textbox(
|
106 |
+
show_label=False,
|
107 |
+
placeholder="Enter text and press enter",
|
108 |
+
).style(container=False)
|
109 |
+
|
110 |
+
with gr.Column(scale=0.10):
|
111 |
+
submit_btn = gr.Button(
|
112 |
+
'Submit',
|
113 |
+
variant='primary'
|
114 |
+
)
|
115 |
+
|
116 |
+
with gr.Column(scale=0.10):
|
117 |
+
clear_btn = gr.Button(
|
118 |
+
'Clear',
|
119 |
+
variant='stop'
|
120 |
+
)
|
121 |
+
# Tanto el submit (hacer enter en el campo de texto) como el submit_btn hacen la misma accion
|
122 |
+
txt.submit(fn=add_text, inputs=[chatbot, txt], outputs=[chatbot, txt] # Cuando envio el submit hago esta funcion
|
123 |
+
).then(fn=bot, inputs=chatbot, outputs=chatbot) # Luego hago esta otra funcion
|
124 |
+
|
125 |
+
submit_btn.click(fn=add_text, inputs=[chatbot, txt], outputs=[chatbot, txt]
|
126 |
+
).then(fn=bot, inputs=chatbot, outputs=chatbot)
|
127 |
+
|
128 |
+
clear_btn.click(lambda: None, None, chatbot, queue=False)
|
129 |
+
|
130 |
+
api_key_button.click(fn=process_key, inputs=[open_ai_key], outputs=None)
|
131 |
+
|
132 |
+
demo.launch(inline=False)
|
requirements.txt
ADDED
@@ -0,0 +1,155 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
accelerate @ git+https://github.com/huggingface/accelerate.git@109f3272f542cbd0e34022f5455078a4ab99d7eb
|
2 |
+
aiofiles==23.1.0
|
3 |
+
aiohttp==3.8.4
|
4 |
+
aiosignal==1.3.1
|
5 |
+
altair==5.0.1
|
6 |
+
anyio==3.7.0
|
7 |
+
arxiv==1.4.7
|
8 |
+
asttokens==2.2.1
|
9 |
+
async-timeout==4.0.2
|
10 |
+
attrs==23.1.0
|
11 |
+
backcall==0.2.0
|
12 |
+
backoff==2.2.1
|
13 |
+
bitsandbytes @ https://github.com/acpopescu/bitsandbytes/releases/download/v0.38.0-win0/bitsandbytes-0.38.1-py3-none-any.whl
|
14 |
+
certifi==2023.5.7
|
15 |
+
charset-normalizer==3.1.0
|
16 |
+
chromadb==0.3.25
|
17 |
+
click==8.1.3
|
18 |
+
clickhouse-connect==0.5.25
|
19 |
+
colorama==0.4.6
|
20 |
+
coloredlogs==15.0.1
|
21 |
+
comm==0.1.3
|
22 |
+
contourpy==1.0.7
|
23 |
+
cycler==0.11.0
|
24 |
+
dataclasses-json==0.5.7
|
25 |
+
debugpy==1.6.7
|
26 |
+
decorator==5.1.1
|
27 |
+
diffusers==0.16.1
|
28 |
+
duckdb==0.8.0
|
29 |
+
einops==0.6.1
|
30 |
+
exceptiongroup==1.1.1
|
31 |
+
executing==1.2.0
|
32 |
+
fastapi==0.96.0
|
33 |
+
feedparser==6.0.10
|
34 |
+
ffmpy==0.3.0
|
35 |
+
filelock==3.12.0
|
36 |
+
flatbuffers==23.5.26
|
37 |
+
fonttools==4.39.4
|
38 |
+
frozenlist==1.3.3
|
39 |
+
fsspec==2023.5.0
|
40 |
+
gradio==3.34.0
|
41 |
+
gradio_client==0.2.6
|
42 |
+
greenlet==2.0.2
|
43 |
+
h11==0.14.0
|
44 |
+
hnswlib==0.7.0
|
45 |
+
httpcore==0.17.2
|
46 |
+
httptools==0.5.0
|
47 |
+
httpx==0.24.1
|
48 |
+
huggingface-hub==0.15.1
|
49 |
+
humanfriendly==10.0
|
50 |
+
idna==3.4
|
51 |
+
importlib-metadata==6.6.0
|
52 |
+
ipykernel==6.23.1
|
53 |
+
ipython==8.14.0
|
54 |
+
ipywidgets==8.0.6
|
55 |
+
jedi==0.18.2
|
56 |
+
Jinja2==3.1.2
|
57 |
+
joblib==1.2.0
|
58 |
+
jsonschema==4.17.3
|
59 |
+
jupyter_client==8.2.0
|
60 |
+
jupyter_core==5.3.0
|
61 |
+
jupyterlab-widgets==3.0.7
|
62 |
+
kiwisolver==1.4.4
|
63 |
+
langchain==0.0.189
|
64 |
+
linkify-it-py==2.0.2
|
65 |
+
lz4==4.3.2
|
66 |
+
markdown-it-py==2.2.0
|
67 |
+
MarkupSafe==2.1.2
|
68 |
+
marshmallow==3.19.0
|
69 |
+
marshmallow-enum==1.5.1
|
70 |
+
matplotlib==3.7.1
|
71 |
+
matplotlib-inline==0.1.6
|
72 |
+
mdit-py-plugins==0.3.3
|
73 |
+
mdurl==0.1.2
|
74 |
+
monotonic==1.6
|
75 |
+
mpmath==1.3.0
|
76 |
+
multidict==6.0.4
|
77 |
+
mypy-extensions==1.0.0
|
78 |
+
nest-asyncio==1.5.6
|
79 |
+
networkx==3.1
|
80 |
+
nltk==3.8.1
|
81 |
+
numexpr==2.8.4
|
82 |
+
numpy==1.24.3
|
83 |
+
onnxruntime==1.15.0
|
84 |
+
openai==0.27.7
|
85 |
+
openapi-schema-pydantic==1.2.4
|
86 |
+
orjson==3.9.0
|
87 |
+
overrides==7.3.1
|
88 |
+
packaging==23.1
|
89 |
+
pandas==2.0.2
|
90 |
+
parso==0.8.3
|
91 |
+
peft @ git+https://github.com/huggingface/peft.git@fcff23f005fc7bfb816ad1f55360442c170cd5f5
|
92 |
+
pickleshare==0.7.5
|
93 |
+
Pillow==9.3.0
|
94 |
+
platformdirs==3.5.1
|
95 |
+
posthog==3.0.1
|
96 |
+
prompt-toolkit==3.0.38
|
97 |
+
protobuf==4.23.2
|
98 |
+
psutil==5.9.5
|
99 |
+
pure-eval==0.2.2
|
100 |
+
pydantic==1.10.8
|
101 |
+
pydub==0.25.1
|
102 |
+
Pygments==2.15.1
|
103 |
+
pyodbc==4.0.39
|
104 |
+
pyparsing==3.0.9
|
105 |
+
pypdf==3.9.1
|
106 |
+
pyreadline3==3.4.1
|
107 |
+
pyrsistent==0.19.3
|
108 |
+
python-dateutil==2.8.2
|
109 |
+
python-dotenv==1.0.0
|
110 |
+
python-multipart==0.0.6
|
111 |
+
pytomlpp @ file:///D:/projects/proy__sql_langchain/pytomlpp-1.0.11-cp310-cp310-win_amd64.whl
|
112 |
+
pytz==2023.3
|
113 |
+
pywin32==306
|
114 |
+
PyYAML==6.0
|
115 |
+
pyzmq==25.1.0
|
116 |
+
regex==2023.5.5
|
117 |
+
requests==2.31.0
|
118 |
+
safetensors==0.3.1
|
119 |
+
scikit-learn==1.2.2
|
120 |
+
scipy==1.10.1
|
121 |
+
semantic-version==2.10.0
|
122 |
+
sentence-transformers==2.2.2
|
123 |
+
sentencepiece==0.1.99
|
124 |
+
sgmllib3k==1.0.0
|
125 |
+
six==1.16.0
|
126 |
+
sniffio==1.3.0
|
127 |
+
SQLAlchemy==2.0.15
|
128 |
+
stack-data==0.6.2
|
129 |
+
starlette==0.27.0
|
130 |
+
sympy==1.12
|
131 |
+
tenacity==8.2.2
|
132 |
+
text-generation==0.6.0
|
133 |
+
threadpoolctl==3.1.0
|
134 |
+
tiktoken==0.4.0
|
135 |
+
tokenizers==0.13.3
|
136 |
+
toolz==0.12.0
|
137 |
+
torch==2.0.1+cu117
|
138 |
+
torchvision==0.15.2
|
139 |
+
tornado==6.3.2
|
140 |
+
tqdm==4.65.0
|
141 |
+
traitlets==5.9.0
|
142 |
+
transformers @ git+https://github.com/huggingface/transformers.git@bacaab1629972b85664fe61ec3caa4da7b55b041
|
143 |
+
typing-inspect==0.9.0
|
144 |
+
typing_extensions==4.6.3
|
145 |
+
tzdata==2023.3
|
146 |
+
uc-micro-py==1.0.2
|
147 |
+
urllib3==2.0.2
|
148 |
+
uvicorn==0.22.0
|
149 |
+
watchfiles==0.19.0
|
150 |
+
wcwidth==0.2.6
|
151 |
+
websockets==11.0.3
|
152 |
+
widgetsnbextension==4.0.7
|
153 |
+
yarl==1.9.2
|
154 |
+
zipp==3.15.0
|
155 |
+
zstandard==0.21.0
|