File size: 20,739 Bytes
a52c7ce c62e5cd 0f7b25d c62e5cd 80b4f00 0f7b25d df62f91 969f5dc 80b4f00 969f5dc c62e5cd 80b4f00 c62e5cd 0f7b25d c62e5cd 0f7b25d 80b4f00 df62f91 cd62201 a835cf0 df62f91 0f7b25d df62f91 0f7b25d a835cf0 df62f91 bc3a2c2 a835cf0 572a6c9 5c15af4 1433ce9 572a6c9 df62f91 1433ce9 572a6c9 1433ce9 969f5dc df62f91 1433ce9 cd62201 1433ce9 cd62201 df62f91 0f7b25d df62f91 55da24c 969f5dc df62f91 0f7b25d df62f91 a835cf0 969f5dc df62f91 0f7b25d df62f91 80b4f00 c62e5cd 969f5dc 55da24c c62e5cd 0f7b25d c62e5cd 0f7b25d a835cf0 572a6c9 a835cf0 80b4f00 a835cf0 80b4f00 a835cf0 0f7b25d 572a6c9 a835cf0 572a6c9 a835cf0 0f7b25d a835cf0 0f7b25d df62f91 0f7b25d 871e8bd 0f7b25d df62f91 0f7b25d df62f91 0f7b25d df62f91 0f7b25d 0f43d97 df62f91 0f7b25d a835cf0 0f7b25d df62f91 a835cf0 572a6c9 a835cf0 df62f91 80b4f00 df62f91 1433ce9 572a6c9 1433ce9 0f7b25d 1433ce9 df62f91 a835cf0 0f7b25d df62f91 0f7b25d c62e5cd |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 |
# chat-pykg/app.py
import datetime
import logging
import os
import random
import shutil
import string
import sys
from pathlib import Path
import numpy as np
import chromadb
import gradio as gr
from chromadb.config import Settings
from langchain.docstore.document import Document
from langchain.embeddings import HuggingFaceEmbeddings, OpenAIEmbeddings
from langchain.vectorstores import Chroma
from chain import get_new_chain1
from ingest import embedding_chooser, ingest_docs
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))
class LogTextboxHandler(logging.StreamHandler):
def __init__(self, textbox):
super().__init__()
self.textbox = textbox
def emit(self, record):
log_entry = self.format(record)
self.textbox.value += f"{log_entry}\n"
def toggle_log_textbox(log_textbox_state):
toggle_visibility = not log_textbox_state
log_textbox_state = not log_textbox_state
return log_textbox_state,gr.update(visible=toggle_visibility)
def update_textbox(full_log):
return gr.update(value=full_log)
def update_radio(radio):
return gr.Radio.update(value=radio)
def change_tab():
return gr.Tabs.update(selected=0)
def update_checkboxgroup(all_collections_state):
new_options = [i for i in all_collections_state]
return gr.CheckboxGroup.update(choices=new_options)
def update_log_textbox(full_log):
return gr.Textbox.update(value=full_log)
def destroy_state(state):
state = None
return state
def clear_chat(chatbot, history):
return [], []
def merge_collections(collection_load_names, vs_state, k_textbox, search_type_selector, vectorstore_radio, embedding_radio):
if type(embedding_radio) == gr.Radio:
embedding_radio = embedding_radio.value
persist_directory = os.path.join(".persisted_data", embedding_radio.replace(' ','_'))
persist_directory_raw = Path('.persisted_data_raw')
embedding_function = embedding_chooser(embedding_radio)
merged_documents = []
merged_embeddings = []
merged_vectorstore = None
if vectorstore_radio == 'Chroma':
for collection_name in collection_load_names:
chroma_obj_get = chromadb.Client(Settings(
chroma_db_impl="duckdb+parquet",
persist_directory=persist_directory,
anonymized_telemetry = True
))
if collection_name == '':
continue
collection_obj = chroma_obj_get.get_collection(collection_name, embedding_function=embedding_function)
collection = collection_obj.get(include=["metadatas", "documents", "embeddings"])
for i in range(len(collection['documents'])):
merged_documents.append(Document(page_content=collection['documents'][i], metadata = collection['metadatas'][i]))
merged_embeddings.append(collection['embeddings'][i])
merged_vectorstore = Chroma(collection_name="temp", embedding_function=embedding_function)
merged_vectorstore.add_documents(documents=merged_documents, embeddings=merged_embeddings)
if vectorstore_radio == 'raw':
merged_vectorstore = []
for collection_name in collection_load_names:
if collection_name == '':
continue
collection_path = persist_directory_raw / collection_name
docarr = np.load(collection_path.as_posix() +'.npy', allow_pickle=True)
merged_vectorstore.extend(docarr.tolist())
# read every line and append to texts
# for f in os.listdir(collection_path):
# with open(os.path.join(collection_path, f), "r") as f:
# merged_vectorstore.append(f.readlines())
return merged_vectorstore
def set_chain_up(openai_api_key, google_api_key, google_cse_id, model_selector, k_textbox, search_type_selector, max_tokens_textbox, vectorstore_radio, vectorstore, agent):
if not agent or type(agent) == str:
if vectorstore != None:
if model_selector in ["gpt-3.5-turbo", "gpt-4"]:
if openai_api_key:
os.environ["OPENAI_API_KEY"] = openai_api_key
os.environ["GOOGLE_API_KEY"] = google_api_key
os.environ["GOOGLE_CSE_ID"] = google_cse_id
qa_chain = get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, search_type_selector, max_tokens_textbox)
os.environ["OPENAI_API_KEY"] = ""
os.environ["GOOGLE_API_KEY"] = ""
os.environ["GOOGLE_CSE_ID"] = ""
return qa_chain
else:
return 'no_open_aikey'
else:
qa_chain = get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, search_type_selector, max_tokens_textbox)
return qa_chain
else:
return 'no_vectorstore'
else:
return agent
def delete_collection(all_collections_state, collections_viewer, select_vectorstore_radio, embedding_radio):
if type(embedding_radio) == gr.Radio:
embedding_radio = embedding_radio.value
persist_directory = os.path.join(".persisted_data", embedding_radio.replace(' ','_'))
persist_directory_raw = Path('.persisted_data_raw')
if select_vectorstore_radio == 'Chroma':
client = chromadb.Client(Settings(
chroma_db_impl="duckdb+parquet",
persist_directory=persist_directory # Optional, defaults to .chromadb/ in the current directory
))
for collection in collections_viewer:
try:
client.delete_collection(collection)
all_collections_state.remove(collection)
collections_viewer.remove(collection)
except Exception as e:
logging.error(e)
if select_vectorstore_radio == 'raw':
for collection in collections_viewer:
try:
os.remove(os.path.join(persist_directory_raw.as_posix(), collection+'.npy' ))
all_collections_state.remove(collection)
collections_viewer.remove(collection)
except Exception as e:
logging.error(e)
return all_collections_state, collections_viewer
def delete_all_collections(all_collections_state, select_vectorstore_radio, embedding_radio):
if type(embedding_radio) == gr.Radio:
embedding_radio = embedding_radio.value
persist_directory = os.path.join(".persisted_data", embedding_radio.replace(' ','_'))
persist_directory_raw = Path('.persisted_data_raw')
if select_vectorstore_radio == 'Chroma':
shutil.rmtree(persist_directory)
if select_vectorstore_radio == 'raw':
shutil.rmtree(persist_directory_raw)
return []
def list_collections(all_collections_state, select_vectorstore_radio, embedding_radio):
if type(embedding_radio) == gr.Radio:
embedding_radio = embedding_radio.value
persist_directory = os.path.join(".persisted_data", embedding_radio.replace(' ','_'))
persist_directory_raw = Path('.persisted_data_raw')
if select_vectorstore_radio == 'Chroma':
client = chromadb.Client(Settings(
chroma_db_impl="duckdb+parquet",
persist_directory=persist_directory # Optional, defaults to .chromadb/ in the current directory
))
collection_names = [[c.name][0] for c in client.list_collections()]
return collection_names
if select_vectorstore_radio == 'raw':
if os.path.exists(persist_directory_raw):
return [f.name.split('.npy')[0] for f in os.scandir(persist_directory_raw)]
return []
def chat(inp, history, agent):
history = history or []
if type(agent) == str:
if agent == 'no_open_aikey':
history.append((inp, "Please paste your OpenAI key to use"))
return history, history
if agent == 'no_vectorstore':
history.append((inp, "Please ingest some package docs to use"))
return history, history
else:
print("\n==== date/time: " + str(datetime.datetime.now()) + " ====")
print("inp: " + inp)
history = history or []
output = agent({"question": inp, "chat_history": history})
answer = output["answer"]
history.append((inp, answer))
print(history)
return history, history
block = gr.Blocks(title = "chat-pykg", analytics_enabled = False, css=".gradio-container {background-color: system;}")
with block:
gr.Markdown("<h1><center>chat-pykg</center></h1>")
with gr.Tabs() as tabs:
with gr.TabItem("Chat", id=0):
with gr.Row():
chatbot = gr.Chatbot()
with gr.Row():
clear_btn = gr.Button("Clear Chat", variant="secondary").style(full_width=False)
message = gr.Textbox(
label="What's your question?",
placeholder="What is this code?",
lines=1,
)
submit = gr.Button(value="Send").style(full_width=False)
gr.Examples(
examples=[
"I want to change the chat-pykg app to have a log viewer, where the user can see what python is doing in the background. How could I do that?",
"Hello, I want to allow chat-pykg to search google before answering. In the langchain docs it says you can use a tool to do this: from langchain.agents import load_tools\ntools = load_tools([“google-search”]). How would I need to change get_new_chain1 function to use tools when it needs to as well as searching the vectorstore? Thanks!",
"Great, thanks. What if I want to add other tools in the future? Can you please change get_new_chain1 function to do that?"
],
inputs=message,
)
with gr.Row():
with gr.Column(scale=1):
model_selector = gr.Dropdown(
choices=["gpt-3.5-turbo", "gpt-4", "other"],
label="Model",
show_label=True,
value = "gpt-4"
)
k_textbox = gr.Textbox(
placeholder="k: Number of search results to consider",
label="Search Results k:",
show_label=True,
lines=1,
value="20",
)
search_type_selector = gr.Dropdown(
choices=["similarity", "mmr", "svm"],
label="Search Type",
show_label=True,
value = "similarity"
)
max_tokens_textbox = gr.Textbox(
placeholder="max_tokens: Maximum number of tokens to generate",
label="max_tokens",
show_label=True,
lines=1,
value="500",
)
with gr.Column(scale=1):
openai_api_key_textbox = gr.Textbox(
placeholder="Paste your OpenAI API key (sk-...)",
show_label=True,
lines=1,
type="password",
label="OpenAI API Key",
)
google_api_key_textbox = gr.Textbox(
placeholder="Paste your Google API key (AIza...)",
show_label=True,
lines=1,
type="password",
label="Google API Key",
)
google_cse_id_textbox = gr.Textbox(
placeholder="Paste your Google CSE ID (0123...)",
show_label=True,
lines=1,
type="password",
label="Google CSE ID",
)
gr.HTML(
"""
This simple application is an implementation of ChatGPT but over an external dataset.
The source code is split/broken down into many document objects using langchain's pythoncodetextsplitter, which apparently tries to keep whole functions etc. together. This means that each file in the source code is split into many smaller documents, and the k value is the number of documents to consider when searching for the most similar documents to the question. With gpt-3.5-turbo, k=10 seems to work well, but with gpt-4, k=20 seems to work better.
The model's memory is set to 5 messages, but I haven't tested with gpt-3.5-turbo yet to see if it works well. It seems to work well with gpt-4."""
)
with gr.TabItem("Repository Selector/Manager", id=1):
with gr.Row():
collections_viewer = gr.CheckboxGroup(choices=[], label='Repository Viewer', show_label=True)
with gr.Row():
load_collections_button = gr.Button(value="Load respositories to chat!", variant="secondary")#.style(full_width=False)
get_all_collection_names_button = gr.Button(value="List all saved repositories", variant="secondary")#.style(full_width=False)
delete_collections_button = gr.Button(value="Delete selected saved repositories", variant="secondary")#.style(full_width=False)
delete_all_collections_button = gr.Button(value="Delete all saved repositories", variant="secondary")#.style(full_width=False)
with gr.Row():
select_embedding_radio = gr.Radio(
choices = ['Sentence Transformers', 'OpenAI'],
label="Embedding Options",
show_label=True,
value='Sentence Transformers'
)
select_vectorstore_radio = gr.Radio(
choices = ['Chroma', 'raw'],
label="Vectorstore Options",
show_label=True,
value='Chroma'
)
with gr.TabItem("Get New Repositories", id=2):
with gr.Row():
all_collections_to_get = gr.List(headers=['Repository (organisation/repo_name)', 'Folders (folder1,folder2...)'], row_count=3, col_count=2, label='Repositories to get', show_label=True, interactive=True, max_cols=2, max_rows=3)
make_collections_button = gr.Button(value="Get new repositories", variant="secondary").style(full_width=False)
with gr.Row():
chunk_size_textbox = gr.Textbox(
placeholder="Chunk size",
label="Chunk size",
show_label=True,
lines=1,
value="2000"
)
chunk_overlap_textbox = gr.Textbox(
placeholder="Chunk overlap",
label="Chunk overlap",
show_label=True,
lines=1,
value="200"
)
make_embedding_radio = gr.Radio(
choices = ['Sentence Transformers', 'OpenAI'],
label="Embedding Options",
show_label=True,
value='Sentence Transformers'
)
make_vectorstore_radio = gr.Radio(
choices = ['Chroma', 'raw'],
label="Vectorstore Options",
show_label=True,
value='Chroma'
)
with gr.Row():
gr.HTML('<center>See the <a href=https://python.langchain.com/en/latest/reference/modules/text_splitter.html>Langchain textsplitter docs</a></center>')
history_state = gr.State()
agent_state = gr.State()
vs_state = gr.State()
all_collections_state = gr.State()
chat_state = gr.State()
debug_state = gr.State()
debug_state.value = False
radio_state = gr.State()
submit.click(set_chain_up, inputs=[openai_api_key_textbox, google_api_key_textbox, google_cse_id_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, vs_state, agent_state], outputs=[agent_state]).then(chat, inputs=[message, history_state, agent_state], outputs=[chatbot, history_state])
message.submit(set_chain_up, inputs=[openai_api_key_textbox, google_api_key_textbox, google_cse_id_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, vs_state, agent_state], outputs=[agent_state]).then(chat, inputs=[message, history_state, agent_state], outputs=[chatbot, history_state])
load_collections_button.click(merge_collections, inputs=[collections_viewer, vs_state, k_textbox, search_type_selector, select_vectorstore_radio, select_embedding_radio], outputs=[vs_state])#.then(change_tab, None, tabs) #.then(set_chain_up, inputs=[openai_api_key_textbox, model_selector, k_textbox, max_tokens_textbox, vs_state, agent_state], outputs=[agent_state])
make_collections_button.click(ingest_docs, inputs=[all_collections_state, all_collections_to_get, chunk_size_textbox, chunk_overlap_textbox, select_vectorstore_radio, select_embedding_radio, debug_state], outputs=[all_collections_state, all_collections_to_get], show_progress=True).then(update_checkboxgroup, inputs = [all_collections_state], outputs = [collections_viewer])
delete_collections_button.click(delete_collection, inputs=[all_collections_state, collections_viewer, select_vectorstore_radio, select_embedding_radio], outputs=[all_collections_state, collections_viewer]).then(update_checkboxgroup, inputs = [all_collections_state], outputs = [collections_viewer])
delete_all_collections_button.click(delete_all_collections, inputs=[all_collections_state,select_vectorstore_radio, select_embedding_radio], outputs=[all_collections_state]).then(update_checkboxgroup, inputs = [all_collections_state], outputs = [collections_viewer])
get_all_collection_names_button.click(list_collections, inputs=[all_collections_state,select_vectorstore_radio, select_embedding_radio], outputs=[all_collections_state]).then(update_checkboxgroup, inputs = [all_collections_state], outputs = [collections_viewer])
clear_btn.click(clear_chat, inputs = [chatbot, history_state], outputs = [chatbot, history_state])
make_embedding_radio.change(update_radio, inputs = make_embedding_radio, outputs = select_embedding_radio)
select_embedding_radio.change(update_radio, inputs = select_embedding_radio, outputs = make_embedding_radio)
make_vectorstore_radio.change(update_radio, inputs =make_vectorstore_radio, outputs = select_vectorstore_radio)
select_vectorstore_radio.change(update_radio, inputs = select_vectorstore_radio, outputs = make_vectorstore_radio)
# Whenever chain parameters change, destroy the agent.
input_list = [openai_api_key_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, make_embedding_radio]
output_list = [agent_state]
for input_item in input_list:
input_item.change(
destroy_state,
inputs=output_list,
outputs=output_list,
)
all_collections_state.value = list_collections(all_collections_state, select_vectorstore_radio, select_embedding_radio)
block.load(update_checkboxgroup, inputs = all_collections_state, outputs = collections_viewer)
log_textbox_handler = LogTextboxHandler(gr.TextArea(interactive=False, placeholder="Logs will appear here...", visible=False))
log_textbox = log_textbox_handler.textbox
logging.getLogger().addHandler(log_textbox_handler)
log_textbox_visibility_state = gr.State()
log_textbox_visibility_state.value = False
log_toggle_button = gr.Button("Toggle Log", variant="secondary")
log_toggle_button.click(toggle_log_textbox, inputs=[log_textbox_visibility_state], outputs=[log_textbox_visibility_state,log_textbox])
gr.HTML(
"<center>Powered by <a href='https://github.com/hwchase17/langchain'>LangChain 🦜️🔗</a></center>"
)
block.queue(concurrency_count=40)
block.launch(debug=True)
|