File size: 38,769 Bytes
a267b49 44b8cfd fdac275 03c7545 a267b49 9cd87f4 f57dc09 815b38b 79c013d fe3d214 ee34418 5e52199 16dc9e5 fdc7196 15d08bb a267b49 ab422ce a267b49 215b4f6 a267b49 9cd87f4 4f26794 8b5cd21 6fe701f 9cd87f4 7831057 be0c4d6 4966e90 ce518a9 b4676f8 1fad9e0 346e521 ee34418 50c35e3 15d08bb ee34418 9cd87f4 2ab448b b40229f 557005f 0cdc7ec 4a5c98c 1e93f85 a9cba95 0cdc7ec ee34418 743bad7 ee34418 836444c ee34418 d88ca48 ee34418 836444c a267b49 6fe701f ee34418 a18030e ee34418 a267b49 8324d7c 69d7f54 f368b42 19a1340 69d7f54 6fe701f 7623b62 6fe701f 0886785 62ab310 3eae162 62ab310 3eae162 62ab310 3eae162 0c81177 7831057 0c81177 a648bb8 6fe701f 0886785 62ab310 3eae162 62ab310 3eae162 62ab310 3eae162 8324d7c 62ab310 854bce1 9cc997c c42b490 a267b49 fdc7196 089f309 fdc7196 40162e5 fdc7196 fd4a890 957f5de fd4a890 9131110 fd4a890 346e521 c02020e fd4a890 290fee1 3282011 9cd87f4 a579acb 9cd87f4 a27226c c02020e 9131110 eb42954 3282011 6143b5b 9cd87f4 215effb 9131110 fd4a890 755634f 662b291 fd4a890 a18030e fd4a890 ee34418 fd4a890 ee34418 6143b5b fdc7196 ee34418 fdc7196 6143b5b ee34418 4b4f3b9 fd4a890 731b64e 846607c 6143b5b 8053dc6 828cea7 b256ccf 828cea7 fd4a890 8053dc6 be0a9d4 8053dc6 fd4a890 be0a9d4 6143b5b ee34418 9d0f111 ee34418 9d0f111 ee34418 c78c291 ee34418 ba165fd fdc7196 db4ca19 ee34418 fd4a890 ee34418 fdc7196 c78c291 fdc7196 ee34418 fd4a890 5481353 ee34418 5481353 fdc7196 e64f337 5481353 9702283 9d0f111 5481353 ee34418 fd4a890 755634f fc24904 ee34418 e8532d4 edf7cf6 b8c4816 ee34418 be0a9d4 de6ca3a ee34418 de6ca3a daf786f 9d0f111 daf786f de6ca3a fc24904 be0a9d4 abbb034 073b1bd ff11c95 2c86ea1 ff11c95 2c86ea1 ff11c95 abbb034 fc24904 e8b73cb 420f6f1 755634f c052e54 5088e43 420f6f1 4bceb82 420f6f1 4bceb82 420f6f1 8eb9c8c bda1e9f ebcd496 5088e43 d418a49 420f6f1 ebcd496 a6cb806 85602b3 ced9420 e41def2 85602b3 e41def2 0cb39e5 e41def2 6aa884b 6143b5b bda1e9f c010288 8b7ce86 99b4006 3f440c6 64b0cfb 6aa884b ab422ce dc77bf9 ab422ce dc77bf9 ab422ce 0052cf7 fed2e5f a9cba95 fed2e5f 0052cf7 ab422ce 016a765 0052cf7 5ac1939 0052cf7 359ff38 0052cf7 11c7a3d ab422ce e8b73cb 000aeae afb0db5 f68e408 8b5cd21 afb0db5 60d3cbf d7c4deb e100b25 85714c0 97fdfd3 7cc8887 887fe00 e100b25 5af5765 e100b25 d391b77 97fdfd3 d391b77 41f36e3 049ccb9 d391b77 b027fb2 ea4fe47 d391b77 97fdfd3 fe668d1 97fdfd3 fe668d1 d391b77 97fdfd3 7cc8887 97fdfd3 8b5cd21 6fe701f ed16fc7 6fe701f 8b5cd21 6fe701f 0886785 6fe701f 8b5cd21 1ad1696 fa5807e e8b73cb 1ad1696 4a1c345 c39c941 40bf681 4a1c345 8b5cd21 e100b25 8b5cd21 e8b73cb 8b5cd21 6fe701f e8b73cb bbd0dbd 000aeae 62c49a7 000aeae e8b73cb da6dfaf e100b25 f68e408 822f8d6 f914f00 e8b73cb 86b5d9a 000aeae 48c085d 1b389ed 2c86ea1 1b389ed 48c085d c4b780a 2c86ea1 c4b780a 7f74442 c4b780a 86b5d9a be0a9d4 15d08bb 86b5d9a e8b73cb c4b780a f914f00 dce0b3a 86b5d9a 1d66c65 4f26794 40e37ba f914f00 0cf748c f914f00 b485ba4 f914f00 1014077 f914f00 1014077 f914f00 6aa884b f914f00 6aa884b b9007f7 f914f00 4ccdfbe f914f00 fdc7196 ba165fd fdc7196 f914f00 fdc7196 f914f00 fdc7196 f914f00 93e855d dce0b3a fdc7196 ba165fd fdc7196 e98448e ba165fd fdc7196 ba165fd fdc7196 ba165fd fdc7196 ba165fd fdc7196 ba165fd c9db885 ba165fd f914f00 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 |
import os
import re
import lz4
import copy
import time
import vllm
import uuid
import torch
import spacy
import shutil
import msgpack
import tempfile
import threading
import itertools
import validators
import numpy as np
import pandas as pd
import streamlit as st
from pathlib import Path
from numpy import ndarray
from outlines import models
from datetime import datetime
from typing import List, Dict
from transformers import AutoTokenizer
from qdrant_client import QdrantClient
from optimum_encoder import OptimumEncoder
from streamlit_navigation_bar import st_navbar
from ppt_chunker import ppt_chunk
from unstructured.cleaners.core import clean
from unstructured.partition.pptx import partition_pptx
from fastembed import SparseTextEmbedding, SparseEmbedding
from unstructured.nlp.tokenize import download_nltk_packages
from huggingface_hub import snapshot_download, hf_hub_download
from scipy.sparse import csr_matrix, save_npz, load_npz, vstack
from langchain_experimental.text_splitter import SemanticChunker
from langchain_community.document_loaders import WikipediaLoader, WebBaseLoader
from qdrant_client.models import (
NamedSparseVector,
NamedVector,
SparseVector,
PointStruct,
ScoredPoint,
Prefetch,
FusionQuery,
Fusion,
SearchRequest,
Modifier,
OptimizersConfigDiff,
HnswConfigDiff,
Distance,
VectorParams,
SparseVectorParams,
SparseIndexParams,
Batch
)
icon_to_types = {
'ppt':('data:image/png;base64,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',
'Powerpoint'),
'pptx':('data:image/png;base64,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',
'Powerpoint'),
'txt':('data:image/png;base64,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',
'Txt'),
'doc':('data:image/png;base64,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',
'Microsoft Word'),
'docx':('data:image/png;base64,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',
'Microsoft Word'),
'xslx':('data:image/png;base64,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',
'Excel')
}
def transform_query(query: str) -> str:
""" For retrieval, add the prompt for query (not for documents).
"""
return f'Represent this sentence for searching relevant passages: {query}'
def query_hybrid_search(query: str, client: QdrantClient, collection_name: str, dense_model: OptimumEncoder, sparse_model: SparseTextEmbedding):
dense_embeddings = dense_model.embed_query(transform_query(query))[0]
sparse_embeddings = list(sparse_model.query_embed(query))[0]
return client.query_points(
collection_name=collection_name,
prefetch=[
Prefetch(query=sparse_embeddings.as_object(), using="text-sparse", limit=10),
Prefetch(query=dense_embeddings, using="text-dense", limit=10)
],
query=FusionQuery(fusion=Fusion.RRF),
limit=3
)
def main(query: str, client: QdrantClient, collection_name: str, tokenizer: AutoTokenizer, llm: vllm.LLM, dense_model: OptimumEncoder, sparse_model: SparseTextEmbedding):
scored_points = query_hybrid_search(query, client, collection_name, dense_model, sparse_model).points
docs = [(scored_point.payload['text'], scored_point.payload['metadata']) for scored_point in scored_points]
contents, metadatas = [list(t) for t in zip(*docs)]
context = "\n".join(contents)
seen_values = set()
result_metadatas = "\n\n".join(
f'{value}'
for metadata in metadatas
for key, value in metadata.items()
if (value not in seen_values and not seen_values.add(value))
)
args = {'context': context, 'query': query}
messages = [
{"role": "system", "content": 'You are a helpful assistant.'},
{"role": "user", "content": st.session_state.toggle_docs['qa_prompt'].format(**args)}
]
prompts = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = llm.generate(
prompts=prompts,
sampling_params=vllm.SamplingParams(
temperature=0,
max_tokens=3000
)
)
print(f'TEXT: {outputs}')
text = outputs[0].outputs[0].text
messages_2 = [
{"role": "system", "content": """Act like a professional summary writer. You have been providing summarization services for various types of documents, including academic papers, legal texts, and business reports, for over 20 years.
Your expertise includes extracting key points and important details concisely without adding unnecessary introductory phrases."""
},
{"role": "user", "content": f"""Write a summary of the following text delimited by triple backquotes. Ensure the summary covers the key points of the text. Do not introduce the summary with sentences like "Here is the summary:" or similar. The summary should be detailed, precise, and directly convey the essential information from the text.
```{text}```
Let's think step-by-step."""
}
]
prompts_2 = tokenizer.apply_chat_template(messages_2, tokenize=False, add_generation_prompt=True)
outputs_2 = llm.generate(
prompts=prompts_2,
sampling_params=vllm.SamplingParams(
temperature=0.3,
max_tokens=3000
)
)
answer = outputs_2[0].outputs[0].text
answer_with_metadatas = f"{answer}\n\n\nSource(s) :\n\n{result_metadatas}"
if st.session_state.documents_only:
return answer if 'no_answer' in text else answer_with_metadatas
else:
return f'Internal Knowledge :\n\n{answer}' if 'knowledge_topic' in text else f'Documents Based :\n\n{answer_with_metadatas}'
def collect_files(directory, pattern):
array = []
for filename in os.listdir(directory):
if pattern in filename:
if filename.endswith('.msgpack'):
with open(os.path.join(directory, filename), "rb") as data_file_payload:
decompressed_payload = data_file_payload.read()
array.append(msgpack.unpackb(decompressed_payload, raw=False))
elif (filename.endswith('.npz')) and (pattern == '_dense'):
v = list(np.load(os.path.join(directory, filename)).values())
print(f'check {v}, {type(v)}, {type(v[0])}')
array.append(v)
elif (filename.endswith('.npz')) and (pattern == '_sparse'):
sparse_embeddings = []
loaded_sparse_matrix = load_npz(os.path.join(directory, filename))
for i in range(loaded_sparse_matrix.shape[0]):
row = loaded_sparse_matrix.getrow(i)
values = row.data.tolist()
indices = row.indices.tolist()
embedding = SparseVector(indices=indices, values=values)
sparse_embeddings.append(embedding)
array.append(sparse_embeddings)
elif filename.endswith('.npy'):
array.append(np.load(os.path.join(directory, filename), allow_pickle=True).tolist())
return list(itertools.chain.from_iterable(array))
@st.cache_resource(show_spinner=False)
def load_models_and_documents():
container = st.empty()
with container.status("Load AI Models and Prepare Documents...", expanded=True) as status:
st.write('Downloading and Loading MixedBread Mxbai Dense Embedding Model under ONNX with Nvidia CUDA as backend...')
dense_model = OptimumEncoder(
device="cuda",
cache_dir=os.getenv('HF_HOME')
)
st.write('Downloading and Loading Qdrant BM42 Sparse Embedding Model under ONNX using the CPU...')
sparse_model = SparseTextEmbedding(
'Qdrant/bm42-all-minilm-l6-v2-attentions',
cache_dir=os.getenv('HF_HOME'),
providers=['CPUExecutionProvider']
)
st.write('Downloading Mistral Nemo AI Model...')
model_path = snapshot_download('casperhansen/mistral-nemo-instruct-2407-awq')
st.write('Loading Mistral Nemo AI Model quantized with AWQ and using Outlines + vLLM Engine as backend...')
tokenizer = AutoTokenizer.from_pretrained(model_path)
llm = vllm.LLM(
model=model_path,
tensor_parallel_size=1,
trust_remote_code=True,
enforce_eager=True,
quantization="awq",
gpu_memory_utilization=1,
max_model_len=12288,
dtype=torch.float16
)
model = models.VLLM(llm)
st.write('Loading Spacy Natural Language Processing Model for English...')
nlp = spacy.load("en_core_web_sm")
download_nltk_packages()
st.write('Creating Collection for our Qdrant Vector Database in RAM memory...')
client = QdrantClient(':memory:')
collection_name = 'collection_demo'
client.create_collection(
collection_name,
{
"text-dense": VectorParams(
size=1024,
distance=Distance.COSINE,
on_disk=False
)
},
{
"text-sparse": SparseVectorParams(
index=SparseIndexParams(
on_disk=False
),
modifier=Modifier.IDF
)
},
2,
optimizers_config=OptimizersConfigDiff(
indexing_threshold=0,
default_segment_number=4
),
hnsw_config=HnswConfigDiff(
on_disk=False,
m=32,
ef_construct=200
)
)
name = 'action_rpg'
embeddings_path = os.path.join(os.getenv('HF_HOME'), 'embeddings')
payload_path = os.path.join(embeddings_path, name + '_payload.msgpack')
dense_path = os.path.join(embeddings_path, name + '_dense.npz')
sparse_path = os.path.join(embeddings_path, name + '_sparse.npz')
ids_path = os.path.join(embeddings_path, name + '_ids.npy')
if not os.path.exists(embeddings_path):
os.mkdir(embeddings_path)
st.write('Downloading and Loading Video Games Dataset coming from Wikipedia...')
docs_1 = WikipediaLoader(query='Action-RPG').load()
docs_2 = WikipediaLoader(query='Real-time strategy').load()
docs_3 = WikipediaLoader(query='First-person shooter').load()
docs_4 = WikipediaLoader(query='Multiplayer online battle arena').load()
docs_5 = WikipediaLoader(query='List of video game genres').load()
docs = docs_1 + docs_2 + docs_3 + docs_4 + docs_5
texts, metadatas = [], []
for doc in docs:
texts.append(doc.page_content)
del doc.metadata['title']
del doc.metadata['summary']
metadatas.append(doc.metadata)
st.write('Transforming the Wikipedia Video Games Dataset into ingestable format for our Qdrant Vector Database...')
payload_docs, dense_embeddings, sparse_embeddings = chunk_documents(texts, metadatas, dense_model, sparse_model)
st.write('Saving on disk the Wikipedia Video Games Dataset into quickly ingestable format...')
with open(payload_path, "wb") as outfile_texts:
packed_payload = msgpack.packb(payload_docs, use_bin_type=True)
outfile_texts.write(packed_payload)
np.savez_compressed(dense_path, *dense_embeddings)
max_index = 0
for embedding in sparse_embeddings:
if len(embedding.indices) > 0:
max_index = max(max_index, max(embedding.indices))
sparse_matrices = []
for embedding in sparse_embeddings:
data = np.array(embedding.values)
indices = np.array(embedding.indices)
indptr = np.array([0, len(data)])
matrix = csr_matrix((data, indices, indptr), shape=(1, max_index + 1))
sparse_matrices.append(matrix)
combined_sparse_matrix = vstack(sparse_matrices)
save_npz(sparse_path, combined_sparse_matrix)
unique_ids = set()
while len(unique_ids) < len(payload_docs):
unique_ids.add(uuid.uuid4().int)
unique_ids = list(unique_ids)
np.save(ids_path, np.array(unique_ids))
else:
st.write('Loading the saved documents on disk')
patterns = ['_ids', '_payload', '_dense', '_sparse']
extensions = ['.npy', '.msgpack', '.npz']
unique_ids, payload_docs, dense_embeddings, sparse_embeddings = [
collect_files(embeddings_path, pattern) for pattern in patterns
]
st.write('Ingesting saved documents on disk into our Qdrant Vector Database...')
client.upsert(
collection_name,
points=Batch(
ids=unique_ids,
payloads=payload_docs,
vectors={
'text-dense': dense_embeddings,
'text-sparse': sparse_embeddings
}
)
)
client.update_collection(
collection_name=collection_name,
optimizer_config=OptimizersConfigDiff(indexing_threshold=20000)
)
status.update(
label="Processing Complete!", state="complete", expanded=False
)
time.sleep(5)
container.empty()
return client, collection_name, tokenizer, model, llm, dense_model, sparse_model, nlp
def chunk_documents(texts: List[str], metadatas: List[dict], dense_model: OptimumEncoder, sparse_model: SparseTextEmbedding):
text_splitter = SemanticChunker(
dense_model,
breakpoint_threshold_type='standard_deviation'
)
docs = text_splitter.create_documents(texts, metadatas)
payload_docs, documents = [], []
for doc in docs:
payload_docs.append({ 'text': doc.page_content, 'metadata': doc.metadata })
documents.append(doc.page_content)
start_dense = time.time()
dense_embeddings = dense_model.embed_documents(documents)
end_dense = time.time()
final_dense = end_dense - start_dense
print(f'DENSE TIME: {final_dense}')
start_sparse = time.time()
sparse_embeddings = [SparseVector(indices=s.indices.tolist(), values=s.values.tolist()) for s in sparse_model.embed(documents, 32)]
end_sparse = time.time()
final_sparse = end_sparse - start_sparse
print(f'SPARSE TIME: {final_sparse}')
return payload_docs, dense_embeddings, sparse_embeddings
def on_change_documents_only():
if st.session_state.documents_only:
st.session_state.toggle_docs = {
'qa_prompt': """You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, reply with 'no_answer'. Use three sentences maximum and keep the answer concise.
Question: {query}
Context: {context}
Answer:""",
'tooltip': 'The AI answer your questions only considering the documents provided',
'display': True
}
else:
st.session_state.toggle_docs = {
'qa_prompt': """If the context is not relevant, please answer the question by using your own knowledge about the topic.
If you decide to provide information using your own knowledge or general knowledge, write 'knowledge_topic' at the top of your answer.
{context}
Question: {query}""",
'tooltip': """The AI answer your questions considering the documents provided, and if it doesn't found the answer in them, try to find in its own internal knowledge""",
'display': False
}
if __name__ == '__main__':
st.set_page_config(page_title="Multipurpose AI Agent",layout="wide", initial_sidebar_state='collapsed')
client, collection_name, tokenizer, model, llm, dense_model, sparse_model, nlp = load_models_and_documents()
if 'menu_id' not in st.session_state:
st.session_state.menu_id = 'ChatBot'
styles = {
"nav": {
"background-color": "rgb(204, 200, 194)",
},
"div": {
"max-width": "32rem",
},
"span": {
"border-radius": "0.5rem",
"color": "rgb(125, 102, 84)",
"margin": "0 0.125rem",
"padding": "0.4375rem 0.625rem",
},
"active": {
"background-color": "rgba(255, 255, 255, 0.25)",
},
"hover": {
"background-color": "rgba(255, 255, 255, 0.35)",
},
}
st.session_state.menu_id = st_navbar(
['ChatBot', 'Documents'],
st.session_state.menu_id,
options={
'hide_nav': False,
'fix_shadow': False,
'use_padding': False
},
styles=styles
)
st.title('Multipurpose AI Agent')
#st.markdown("<h1 style='position: fixed; top: 0; left: 0; width: 100%; padding: 10px; text-align: left; color: black;'>Multipurpose AI Agent</h1>", unsafe_allow_html=True)
data_editor_path = os.path.join(os.getenv('HF_HOME'), 'documents')
if 'df' not in st.session_state:
if os.path.exists(data_editor_path):
st.session_state.df = pd.read_parquet(os.path.join(data_editor_path, 'data_editor.parquet.sz'), engine='pyarrow')
else:
st.session_state.df = pd.DataFrame()
os.mkdir(data_editor_path)
def on_change_data_editor():
print(f'first key : {st.session_state.key_data_editor}')
print(f'first key : {st.session_state.df}')
if st.session_state.menu_id == 'Documents':
st.session_state.df = st.data_editor(
st.session_state.df,
num_rows="dynamic",
use_container_width=True,
hide_index=True,
on_change=on_change_data_editor,
key='key_data_editor',
column_config={
'icon': st.column_config.ImageColumn(
'Document'
),
"document": st.column_config.TextColumn(
"Name",
help="Name of the document",
required=True
),
"type": st.column_config.SelectboxColumn(
'File type',
help='The file format extension of this document',
required=True,
options=[
'Powerpoint',
'Microsoft Word',
'Excel'
]
),
"path": st.column_config.TextColumn(
'Path',
help='Path to the document',
required=False
),
"time": st.column_config.DatetimeColumn(
'Date and hour',
help='When this document has been ingested here for the last time',
format="D MMM YYYY, h:mm a",
required=True
),
"toggle": st.column_config.CheckboxColumn(
'Enable/Disable',
help='Either to enable or disable the ability for the ai to find this document',
required=True,
default=True
)
}
)
conversations_path = os.path.join(os.getenv('HF_HOME'), 'conversations')
try:
with open(conversations_path, 'rb') as fp:
packed_bytes = fp.read()
conversations: Dict[str, list] = msgpack.unpackb(packed_bytes, raw=False)
except:
conversations = {}
if st.session_state.menu_id == 'ChatBot':
if 'id_chat' not in st.session_state:
st.session_state.id_chat = 'New Chat'
def options_list(conversations: Dict[str, list]):
if st.session_state.id_chat == 'New Chat':
return [st.session_state.id_chat] + list(conversations.keys())
else:
return ['New Chat'] + list(conversations.keys())
with st.sidebar:
st.session_state.id_chat = st.selectbox(
label='Choose a conversation',
options=options_list(conversations),
index=0,
placeholder='_',
key='chat_id'
)
st.session_state.messages = conversations[st.session_state.id_chat] if st.session_state.id_chat != 'New Chat' else []
def update_selectbox_remove(conversations_path, conversations):
conversations.pop(st.session_state.chat_id)
with open(conversations_path, 'wb') as fp:
packed_bytes = msgpack.packb(conversations, use_bin_type=True)
fp.write(packed_bytes)
st.session_state.chat_id = 'New Chat'
st.button(
'Delete Chat',
use_container_width=True,
disabled=False if st.session_state.id_chat != 'New Chat' else True,
on_click=update_selectbox_remove,
args=(conversations_path, conversations)
)
def generate_conv_title(llm: vllm.LLM, tokenizer: AutoTokenizer):
if st.session_state.chat_id == 'New Chat':
messages = [
{"role": "system", "content": 'You are a helpful assistant.'},
{"role": "user", "content": f"""Understand the question of the user.
Resume in one single sentence what is the subject of the conversation and what is the user talking about.
Question : {st.session_state.user_input}"""
}
]
prompts = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = llm.generate(
prompts=prompts,
sampling_params=vllm.SamplingParams(
temperature=0.3,
max_tokens=30
)
)
st.session_state.chat_id = outputs[0].outputs[0].text
st.session_state.messages = []
conversations.update({st.session_state.chat_id: st.session_state.messages})
with open(conversations_path, 'wb') as fp:
packed_bytes = msgpack.packb(conversations, use_bin_type=True)
fp.write(packed_bytes)
for message in st.session_state.messages:
with st.chat_message(message["role"]):
st.markdown(message["content"])
if prompt := st.chat_input(
"Message Video Game Assistant",
on_submit=generate_conv_title,
key='user_input',
args=(llm, tokenizer)
):
st.chat_message("user").markdown(prompt)
st.session_state.messages.append({"role": "user", "content": prompt})
ai_response = main(prompt, client, collection_name, tokenizer, llm, dense_model, sparse_model)
with st.chat_message("assistant"):
message_placeholder = st.empty()
full_response = ""
for chunk in re.split(r'(\s+)', ai_response):
full_response += chunk + " "
time.sleep(0.2)
message_placeholder.write(full_response + 'β')
message_placeholder.write(re.sub('β', '', full_response))
st.session_state.messages.append({"role": "assistant", "content": full_response})
conversations.update({st.session_state.id_chat: st.session_state.messages})
with open(conversations_path, 'wb') as fp:
packed_bytes = msgpack.packb(conversations, use_bin_type=True)
fp.write(packed_bytes)
if "cached_files" not in st.session_state:
st.session_state.cached_files = []
with st.sidebar:
st.divider()
if 'toggle_docs' not in st.session_state:
st.session_state.toggle_docs = {
'qa_prompt': """You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, reply with 'no_answer'. Use three sentences maximum and keep the answer concise.
Question: {query}
Context: {context}
Answer:""",
'tooltip': 'The AI answer your questions only considering the documents provided',
'display': True
}
st.toggle(
label="""Enable 'Documents-Only' Mode""",
value=st.session_state.toggle_docs['display'],
on_change=on_change_documents_only,
key="documents_only",
help=st.session_state.toggle_docs['tooltip']
)
st.divider()
url = st.text_input("Scrape an URL link :")
if validators.url(url):
docs = WebBaseLoader(url).load()
print(f'WebBaseLoader: {docs[0].metadata}')
texts, metadatas = [], []
for doc in docs:
texts.append(doc.page_content)
del doc.metadata['title']
del doc.metadata['description']
del doc.metadata['language']
metadatas.append(doc.metadata)
payload_docs, dense_embeddings, sparse_embeddings = chunk_documents(texts, metadatas, dense_model, sparse_model)
client.upsert(
collection_name,
make_points(
texts,
metadatas,
dense_embeddings,
sparse_embeddings
)
)
st.toast('URL Content Ingested !', icon='π')
st.divider()
uploaded_files = st.file_uploader("Upload a file :", accept_multiple_files=True, type=['pptx', 'ppt'])
for uploaded_file in uploaded_files:
if uploaded_file not in st.session_state.cached_files:
st.session_state.cached_files.append(uploaded_file)
file_name = os.path.basename(uploaded_file.name)
base_name, ext = os.path.splitext(file_name)
processing_time = datetime.now().strftime('%d %b %Y, %I:%M %p')
full_path = os.path.realpath(uploaded_file.name)
file_type = ext.lstrip('.')
d = {
'icon': icon_to_types[file_type][0],
'document': base_name,
'type': icon_to_types[file_type][1],
'path': full_path,
'time': [datetime.strptime(processing_time, '%d %b %Y, %I:%M %p')],
'toggle': True
}
if (st.session_state.df.empty) or (base_name not in st.session_state.df['document'].tolist()):
st.session_state.df = pd.concat(
[
st.session_state.df,
pd.DataFrame(data={
'icon': icon_to_types[file_type][0],
'document': base_name,
'type': icon_to_types[file_type][1],
'path': full_path,
'time': [datetime.strptime(processing_time, '%d %b %Y, %I:%M %p')],
'toggle': True
})
]
)
else:
idx = st.session_state.df.index[st.session_state.df['document']==base_name].tolist()[0]
st.session_state.df.loc[idx] = {
'icon': icon_to_types[file_type][0],
'document': base_name,
'type': icon_to_types[file_type][1],
'path': full_path,
'time': datetime.strptime(processing_time, '%d %b %Y, %I:%M %p'),
'toggle': True
}
st.session_state.df.to_parquet(
os.path.join(
data_editor_path,
'data_editor.parquet.sz'
),
compression='snappy',
engine='pyarrow'
)
weakDict, tables = ppt_chunk(uploaded_file, nlp)
documents = weakDict.all_texts()
dense = dense_model.embed_documents(documents)
sparse = [s for s in sparse_model.embed(documents, 32)]
ids = weakDict.all_ids()
payload_docs = [{ 'text': documents[i], 'metadata': metadata } for i, metadata in enumerate(weakDict.all_metadatas())]
client.upsert(
collection_name=collection_name,
points=Batch(
ids=ids,
payloads=payload_docs,
vectors={
'text-dense': dense,
'text-sparse': [SparseVector(indices=s.indices.tolist(), values=s.values.tolist()) for s in sparse]
}
)
)
embeddings_path = os.path.join(os.getenv('HF_HOME'), 'embeddings')
payload_path = os.path.join(embeddings_path, base_name + '_payload')
dense_path = os.path.join(embeddings_path, base_name + '_dense.npz')
sparse_path = os.path.join(embeddings_path, base_name + '_sparse.npz')
ids_path = os.path.join(embeddings_path, base_name + '_ids.npy')
with open(payload_path, "wb") as outfile_texts:
packed_payload = msgpack.packb(payload_docs, use_bin_type=True)
outfile_texts.write(packed_payload)
np.savez_compressed(dense_path, *dense)
max_index = 0
for embedding in sparse:
if len(embedding.indices) > 0:
max_index = max(max_index, max(embedding.indices))
sparse_matrices = []
for embedding in sparse:
data = np.array(embedding.values)
indices = np.array(embedding.indices)
indptr = np.array([0, len(data)])
matrix = csr_matrix((data, indices, indptr), shape=(1, max_index + 1))
sparse_matrices.append(matrix)
combined_sparse_matrix = vstack(sparse_matrices)
save_npz(sparse_path, combined_sparse_matrix)
np.save(ids_path, np.array(ids), allow_pickle=True)
st.toast('Document(s) Ingested !', icon='π') |