Spaces:
Running
Running
File size: 37,653 Bytes
885a9dd 7a9ead6 99fac9b 2f7fc69 82089f6 2f7fc69 7a9ead6 2f7fc69 7a9ead6 f557882 885a9dd f557882 885a9dd 6d43bdb 885a9dd 509621c 885a9dd 63a5823 885a9dd e20177b 885a9dd e20177b 509621c 885a9dd |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 |
from io import StringIO
import sys
import os
# Set EasyOCR cache directory to a writable location
os.environ["EASYOCR_CACHE_DIR"] = "/app/.EASYOCR"
import easyocr
# Monkey-patch the easyocr.Reader to force the model_storage directory parameter
_original_init = easyocr.Reader.__init__
def new_init(self, *args, **kwargs):
if args and "lang_list" in kwargs:
del kwargs["lang_list"]
kwargs.setdefault("model_storage_directory", "/app/.EasyOCR")
_original_init(self, *args, **kwargs)
easyocr.Reader.__init__ = new_init
#from huggingface_hub import login
import gradio as gr
import json
import csv
import hashlib
import uuid
import logging
from typing import Annotated, List, Dict, Sequence, TypedDict
# LangChain & related imports
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import tool, StructuredTool
from pydantic import BaseModel, Field
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_chroma import Chroma
from langchain_core.documents import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.retrievers import EnsembleRetriever
# Extraction for Documents
from langchain_docling.loader import ExportType
from langchain_docling import DoclingLoader
from docling.chunking import HybridChunker
# Extraction for HTML
from langchain_community.document_loaders import WebBaseLoader
from urllib.parse import urlparse
from langchain_groq import ChatGroq
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import InjectedStore
from langgraph.store.base import BaseStore
from langgraph.store.memory import InMemoryStore
from langgraph.checkpoint.memory import MemorySaver
from langchain.embeddings import init_embeddings
from langgraph.graph import StateGraph
from langgraph.graph.message import add_messages
from langgraph.prebuilt import ToolNode, tools_condition
from langchain_core.messages import (
SystemMessage,
AIMessage,
HumanMessage,
BaseMessage,
ToolMessage,
)
logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
logger = logging.getLogger(__name__)
# Suppress all library logs at or below WARNING for user experience:
logging.disable(logging.WARNING)
# HF_TOKEN = os.getenv("HF_TOKEN") # Read from environment variable
# if HF_TOKEN:
# login(token=HF_TOKEN) # Log in to Hugging Face Hub
# else:
# print("Warning: HF_TOKEN not found in environment variables.")
GROQ_API_KEY = os.getenv("GROQ_API_KEY") # Read from environment variable
if not GROQ_API_KEY:
print("Warning: GROQ_API_KEY not found in environment variables.")
EMBED_MODEL_ID = "sentence-transformers/all-MiniLM-L6-v2"
# =============================================================================
# Document Extraction Functions
# =============================================================================
def extract_documents(doc_path: str) -> List[str]:
"""
Recursively collects all file paths from folder 'doc_path'.
Used by ExtractDocument.load_files() to find documents to parse.
"""
extracted_docs = []
for root, _, files in os.walk(doc_path):
for file in files:
file_path = os.path.join(root, file)
extracted_docs.append(file_path)
return extracted_docs
def _generate_uuid(page_content: str) -> str:
"""Generate a UUID for a chunk of text using MD5 hashing."""
md5_hash = hashlib.md5(page_content.encode()).hexdigest()
return str(uuid.UUID(md5_hash[0:32]))
def load_file(file_path: str) -> List[Document]:
"""
Load a file from the given path and return a list of Document objects.
"""
_documents = []
# Load the file and extract the text chunks
try:
loader = DoclingLoader(
file_path = file_path,
export_type = ExportType.DOC_CHUNKS,
chunker = HybridChunker(tokenizer=EMBED_MODEL_ID),
)
docs = loader.load()
logger.info(f"Total parsed doc-chunks: {len(docs)} from Source: {file_path}")
for d in docs:
# Tag each document's chunk with the source file and a unique ID
doc = Document(
page_content=d.page_content,
metadata={
"source": file_path,
"doc_id": _generate_uuid(d.page_content),
"source_type": "file",
}
)
_documents.append(doc)
logger.info(f"Total generated LangChain document chunks: {len(_documents)}\n.")
except Exception as e:
logger.error(f"Error loading file: {file_path}. Exception: {e}\n.")
return _documents
# Define function to load documents from a folder
def load_files_from_folder(doc_path: str) -> List[Document]:
"""
Load documents from the given folder path and return a list of Document objects.
"""
_documents = []
# Extract all files path from the given folder
extracted_docs = extract_documents(doc_path)
# Iterate through each document and extract the text chunks
for file_path in extracted_docs:
_documents.extend(load_file(file_path))
return _documents
# =============================================================================
# Load structured data in csv file to LangChain Document format
def load_mcq_csvfiles(file_path: str) -> List[Document]:
"""
Load structured data in mcq csv file from the given file path and return a list of Document object.
Expected format: each row of csv is comma separated into "mcq_number", "mcq_type", "text_content"
"""
_documents = []
# iterate through each csv file and load each row into _dict_per_question format
# Ensure we process only CSV files
if not file_path.endswith(".csv"):
return _documents # Skip non-CSV files
try:
# Open and read the CSV file
with open(file_path, mode='r', encoding='utf-8') as file:
reader = csv.DictReader(file)
for row in reader:
# Ensure required columns exist in the row
if not all(k in row for k in ["mcq_number", "mcq_type", "text_content"]): # Ensure required columns exist and exclude header
logger.error(f"Skipping row due to missing fields: {row}")
continue
# Tag each row of csv is comma separated into "mcq_number", "mcq_type", "text_content"
doc = Document(
page_content = row["text_content"], # text_content segment is separated by "|"
metadata={
"source": f"{file_path}_{row['mcq_number']}", # file_path + mcq_number
"doc_id": _generate_uuid(f"{file_path}_{row['mcq_number']}"), # Unique ID
"source_type": row["mcq_type"], # MCQ type
}
)
_documents.append(doc)
logger.info(f"Successfully loaded {len(_documents)} LangChain document chunks from {file_path}.")
except Exception as e:
logger.error(f"Error loading file: {file_path}. Exception: {e}\n.")
return _documents
# Define function to load documents from a folder for structured data in csv file
def load_files_from_folder_mcq(doc_path: str) -> List[Document]:
"""
Load mcq csv file from the given folder path and return a list of Document objects.
"""
_documents = []
# Extract all files path from the given folder
extracted_docs = [
os.path.join(doc_path, file) for file in os.listdir(doc_path)
if file.endswith(".csv") # Process only CSV files
]
# Iterate through each document and extract the text chunks
for file_path in extracted_docs:
_documents.extend(load_mcq_csvfiles(file_path))
return _documents
# =============================================================================
# Website Extraction Functions
# =============================================================================
def _generate_uuid(page_content: str) -> str:
"""Generate a UUID for a chunk of text using MD5 hashing."""
md5_hash = hashlib.md5(page_content.encode()).hexdigest()
return str(uuid.UUID(md5_hash[0:32]))
def ensure_scheme(url):
parsed_url = urlparse(url)
if not parsed_url.scheme:
return 'http://' + url # Default to http, or use 'https://' if preferred
return url
def extract_html(url: List[str]) -> List[Document]:
if isinstance(url, str):
url = [url]
"""
Extracts text from the HTML content of web pages listed in 'web_path'.
Returns a list of LangChain 'Document' objects.
"""
# Ensure all URLs have a scheme
web_paths = [ensure_scheme(u) for u in url]
loader = WebBaseLoader(web_paths)
loader.requests_per_second = 1
docs = loader.load()
# Iterate through each document, clean the content, removing excessive line return and store it in a LangChain Document
_documents = []
for doc in docs:
# Clean the concent
doc.page_content = doc.page_content.strip()
doc.page_content = doc.page_content.replace("\n", " ")
doc.page_content = doc.page_content.replace("\r", " ")
doc.page_content = doc.page_content.replace("\t", " ")
doc.page_content = doc.page_content.replace(" ", " ")
doc.page_content = doc.page_content.replace(" ", " ")
# Store it in a LangChain Document
web_doc = Document(
page_content=doc.page_content,
metadata={
"source": doc.metadata.get("source"),
"doc_id": _generate_uuid(doc.page_content),
"source_type": "web"
}
)
_documents.append(web_doc)
return _documents
# =============================================================================
# Vector Store Initialisation
# =============================================================================
embedding_model = HuggingFaceEmbeddings(model_name=EMBED_MODEL_ID)
# Initialise vector stores
general_vs = Chroma(
collection_name="general_vstore",
embedding_function=embedding_model,
persist_directory="./general_db"
)
mcq_vs = Chroma(
collection_name="mcq_vstore",
embedding_function=embedding_model,
persist_directory="./mcq_db"
)
in_memory_vs = Chroma(
collection_name="in_memory_vstore",
embedding_function=embedding_model
)
# Split the documents into smaller chunks for better embedding coverage
def split_text_into_chunks(docs: List[Document]) -> List[Document]:
"""
Splits a list of Documents into smaller text chunks using
RecursiveCharacterTextSplitter while preserving metadata.
Returns a list of Document objects.
"""
if not docs:
return []
splitter = RecursiveCharacterTextSplitter(
chunk_size=1000, # Split into chunks of 1000 characters
chunk_overlap=200, # Overlap by 200 characters
add_start_index=True
)
chunked_docs = splitter.split_documents(docs)
return chunked_docs # List of Document objects
# =============================================================================
# Retrieval Tools
# =============================================================================
# Define a simple similarity search retrieval tool on msq_vs
class MCQRetrievalTool(BaseModel):
input: str = Field(..., title="input", description="Search topic.")
k: int = Field(2, title="Number of Results", description="The number of results to retrieve.")
def mcq_retriever(input: str, k: int = 2) -> List[str]:
# Retrieve the top k most similar mcq question documents from the vector store
docs_func = mcq_vs.as_retriever(
search_type="similarity",
search_kwargs={
'k': k,
'filter':{"source_type": "mcq_question"}
},
)
docs_qns = docs_func.invoke(input, k=k)
# Extract the document IDs from the retrieved documents
doc_ids = [d.metadata.get("doc_id") for d in docs_qns if "doc_id" in d.metadata]
# Retrieve full documents based on the doc_ids
docs = mcq_vs.get(where = {'doc_id': {"$in":doc_ids}})
qns_list = {}
for i, d in enumerate(docs['metadatas']):
qns_list[d['source'] + " " + d['source_type']] = docs['documents'][i]
return qns_list
# Create a StructuredTool from the function
mcq_retriever_tool = StructuredTool.from_function(
func = mcq_retriever,
name = "MCQ Retrieval Tool",
description = (
"""
Use this tool to retrieve MCQ questions set when Human asks to generate a quiz related to a topic.
DO NOT GIVE THE ANSWERS to Human before Human has answered all the questions.
If Human give answers for questions you do not know, SAY you do not have the questions for the answer
and ASK if the Human want you to generate a new quiz and then SAVE THE QUIZ with Summary Tool before ending the conversation.
Input must be a JSON string with the schema:
- input (str): The search topic to retrieve MCQ questions set related to the topic.
- k (int): Number of question set to retrieve.
Example usage: input='What is AI?', k=5
Returns:
- A dict of MCQ questions:
Key: 'metadata of question' e.g. './Documents/mcq/mcq.csv_Qn31 mcq_question' with suffix ['question', 'answer', 'answer_reason', 'options', 'wrong_options_reason']
Value: Text Content
"""
),
args_schema = MCQRetrievalTool,
response_format="content",
return_direct = False, # Return the response as a list of strings
verbose = False # To log tool's progress
)
# -----------------------------------------------------------------------------
# Retrieve more documents with higher diversity using MMR (Maximal Marginal Relevance) from the general vector store
# Useful if the dataset has many similar documents
class GenRetrievalTool(BaseModel):
input: str = Field(..., title="input", description="User query.")
k: int = Field(2, title="Number of Results", description="The number of results to retrieve.")
def gen_retriever(input: str, k: int = 2) -> List[str]:
# Use retriever of vector store to retrieve documents
docs_func = general_vs.as_retriever(
search_type="mmr",
search_kwargs = {'k': k, 'lambda_mult': 0.25}
)
docs = docs_func.invoke(input, k=k)
return [d.page_content for d in docs]
# Create a StructuredTool from the function
general_retriever_tool = StructuredTool.from_function(
func = gen_retriever,
name = "Assistant References Retrieval Tool",
description = (
"""
Use this tool to retrieve reference information from Assistant reference database for Human queries related to a topic or
and when Human asked to generate guides to learn or study about a topic.
Input must be a JSON string with the schema:
- input (str): The user query.
- k (int): Number of results to retrieve.
Example usage: input='What is AI?', k=5
Returns:
- A list of retrieved document's content string.
"""
),
args_schema = GenRetrievalTool,
response_format="content",
return_direct = False, # Return the content of the documents
verbose = False # To log tool's progress
)
# -----------------------------------------------------------------------------
# Retrieve more documents with higher diversity using MMR (Maximal Marginal Relevance) from the in-memory vector store
# Query in-memory vector store only
class InMemoryRetrievalTool(BaseModel):
input: str = Field(..., title="input", description="User query.")
k: int = Field(2, title="Number of Results", description="The number of results to retrieve.")
def in_memory_retriever(input: str, k: int = 2) -> List[str]:
# Use retriever of vector store to retrieve documents
docs_func = in_memory_vs.as_retriever(
search_type="mmr",
search_kwargs = {'k': k, 'lambda_mult': 0.25}
)
docs = docs_func.invoke(input, k=k)
return [d.page_content for d in docs]
# Create a StructuredTool from the function
in_memory_retriever_tool = StructuredTool.from_function(
func = in_memory_retriever,
name = "In-Memory Retrieval Tool",
description = (
"""
Use this tool when Human ask Assistant to retrieve information from documents that Human has uploaded.
Input must be a JSON string with the schema:
- input (str): The user query.
- k (int): Number of results to retrieve.
"""
),
args_schema = InMemoryRetrievalTool,
response_format="content",
return_direct = False, # Whether to return the tool’s output directly
verbose = False # To log tool's progress
)
# -----------------------------------------------------------------------------
# Web Extraction Tool
class WebExtractionRequest(BaseModel):
input: str = Field(..., title="input", description="Search text.")
url: str = Field(
...,
title="url",
description="Web URL(s) to extract content from. If multiple URLs, separate them with a comma."
)
k: int = Field(5, title="Number of Results", description="The number of results to retrieve.")
# Extract content from a web URL, load into in_memory_vstore
def extract_web_path_tool(input: str, url: str, k: int = 5) -> List[str]:
if isinstance(url, str):
url = [url]
"""
Extract content from the web URLs based on user's input.
Args:
- input: The input text to search for.
- url: URLs to extract content from.
- k: Number of results to retrieve.
Returns:
- A list of retrieved document's content string.
"""
# Extract content from the web
html_docs = extract_html(url)
if not html_docs:
return f"No content extracted from {url}."
# Split the documents into smaller chunks for better embedding coverage
chunked_texts = split_text_into_chunks(html_docs)
in_memory_vs.add_documents(chunked_texts) # Add the chunked texts to the in-memory vector store
# Extract content from the in-memory vector store
# Use retriever of vector store to retrieve documents
docs_func = in_memory_vs.as_retriever(
search_type="mmr",
search_kwargs={
'k': k,
'lambda_mult': 0.25,
'filter':{"source": {"$in": url}}
},
)
docs = docs_func.invoke(input, k=k)
return [d.page_content for d in docs]
# Create a StructuredTool from the function
web_extraction_tool = StructuredTool.from_function(
func = extract_web_path_tool,
name = "Web Extraction Tool",
description = (
"Assistant should use this tool to extract content from web URLs based on user's input, "
"Web extraction is initially load into database and then return k: Number of results to retrieve"
),
args_schema = WebExtractionRequest,
return_direct = False, # Whether to return the tool’s output directly
verbose = False # To log tool's progress
)
# -----------------------------------------------------------------------------
# Ensemble Retrieval from General and In-Memory Vector Stores
class EnsembleRetrievalTool(BaseModel):
input: str = Field(..., title="input", description="User query.")
k: int = Field(5, title="Number of Results", description="Number of results.")
def ensemble_retriever(input: str, k: int = 5) -> List[str]:
# Use retriever of vector store to retrieve documents
general_retrieval = general_vs.as_retriever(
search_type="mmr",
search_kwargs = {'k': k, 'lambda_mult': 0.25}
)
in_memory_retrieval = in_memory_vs.as_retriever(
search_type="mmr",
search_kwargs = {'k': k, 'lambda_mult': 0.25}
)
ensemble_retriever = EnsembleRetriever(
retrievers=[general_retrieval, in_memory_retrieval],
weights=[0.5, 0.5]
)
docs = ensemble_retriever.invoke(input)
return [d.page_content for d in docs]
# Create a StructuredTool from the function
ensemble_retriever_tool = StructuredTool.from_function(
func = ensemble_retriever,
name = "Ensemble Retriever Tool",
description = (
"""
Use this tool to retrieve information from reference database and
extraction of documents that Human has uploaded.
Input must be a JSON string with the schema:
- input (str): The user query.
- k (int): Number of results to retrieve.
"""
),
args_schema = EnsembleRetrievalTool,
response_format="content",
return_direct = False
)
###############################################################################
# LLM Model Setup
###############################################################################
TEMPERATURE = 0.5
# model = ChatOpenAI(
# model="unsloth/llama-3-8b-Instruct-bnb-4bit",
# temperature=TEMPERATURE,
# timeout=None,
# max_retries=2,
# api_key="not_required",
# base_url="http://localhost:8000/v1", # Use the VLLM instance URL
# verbose=True
# )
model = ChatGroq(
model_name="deepseek-r1-distill-llama-70b",
temperature=TEMPERATURE,
api_key=GROQ_API_KEY,
verbose=True
)
###############################################################################
# 1. Initialize memory + config
###############################################################################
in_memory_store = InMemoryStore(
index={
"embed": init_embeddings("huggingface:sentence-transformers/all-MiniLM-L6-v2"),
"dims": 384, # Embedding dimensions
}
)
# A memory saver to checkpoint conversation states
checkpointer = MemorySaver()
# Initialize config with user & thread info
config = {}
config["configurable"] = {
"user_id": "user_1",
"thread_id": 0,
}
###############################################################################
# 2. Define MessagesState
###############################################################################
class MessagesState(TypedDict):
"""The state of the agent.
The key 'messages' uses add_messages as a reducer,
so each time this state is updated, new messages are appended.
# See https://langchain-ai.github.io/langgraph/concepts/low_level/#reducers
"""
messages: Annotated[Sequence[BaseMessage], add_messages]
###############################################################################
# 3. Memory Tools
###############################################################################
def save_memory(summary_text: str, *, config: RunnableConfig, store: BaseStore) -> str:
"""Save the given memory for the current user and return the key."""
user_id = config.get("configurable", {}).get("user_id")
thread_id = config.get("configurable", {}).get("thread_id")
namespace = (user_id, "memories")
memory_id = thread_id
store.put(namespace, memory_id, {"memory": summary_text})
return f"Saved to memory key: {memory_id}"
def update_memory(state: MessagesState, config: RunnableConfig, *, store: BaseStore):
# Extract the messages list from the event, handling potential missing key
messages = state["messages"]
# Convert LangChain messages to dictionaries before storing
messages_dict = [{"role": msg.type, "content": msg.content} for msg in messages]
# Get the user id from the config
user_id = config.get("configurable", {}).get("user_id")
thread_id = config.get("configurable", {}).get("thread_id")
# Namespace the memory
namespace = (user_id, "memories")
# Create a new memory ID
memory_id = f"{thread_id}"
store.put(namespace, memory_id, {"memory": messages_dict})
return f"Saved to memory key: {memory_id}"
# Define a Pydantic schema for the save_memory tool (if needed elsewhere)
# https://langchain-ai.github.io/langgraphjs/reference/classes/checkpoint.InMemoryStore.html
class RecallMemory(BaseModel):
query_text: str = Field(..., title="Search Text", description="The text to search from memories for similar records.")
k: int = Field(5, title="Number of Results", description="Number of results to retrieve.")
def recall_memory(query_text: str, k: int = 5) -> str:
"""Retrieve user memories from in_memory_store."""
user_id = config.get("configurable", {}).get("user_id")
memories = [
m.value["memory"] for m in in_memory_store.search((user_id, "memories"), query=query_text, limit=k)
if "memory" in m.value
]
return f"User memories: {memories}"
# Create a StructuredTool from the function
recall_memory_tool = StructuredTool.from_function(
func=recall_memory,
name="Recall Memory Tool",
description="""
Retrieve memories relevant to the user's query.
""",
args_schema=RecallMemory,
response_format="content",
return_direct=False,
verbose=False
)
###############################################################################
# 4. Summarize Node (using StructuredTool)
###############################################################################
# Define a Pydantic schema for the Summary tool
class SummariseConversation(BaseModel):
summary_text: str = Field(..., title="text", description="Write a summary of entire conversation here")
def summarise_node(summary_text: str):
"""
Final node that summarizes the entire conversation for the current thread,
saves it in memory, increments the thread_id, and ends the conversation.
Returns a confirmation string.
"""
user_id = config["configurable"]["user_id"]
current_thread_id = config["configurable"]["thread_id"]
new_thread_id = str(int(current_thread_id) + 1)
# Prepare configuration for saving memory with updated thread id
config_for_saving = {
"configurable": {
"user_id": user_id,
"thread_id": new_thread_id
}
}
key = save_memory(summary_text, config=config_for_saving, store=in_memory_store)
#return f"Summary saved under key: {key}"
# Create a StructuredTool from the function (this wraps summarise_node)
summarise_tool = StructuredTool.from_function(
func=summarise_node,
name="Summary Tool",
description="""
Summarize the current conversation in no more than
1000 words. Also retain any unanswered quiz questions along with
your internal answers so the next conversation thread can continue.
Do not reveal solutions to the user yet. Use this tool to save
the current conversation to memory and then end the conversation.
""",
args_schema=SummariseConversation,
response_format="content",
return_direct=False,
verbose=True
)
def call_summary(state: MessagesState, config: RunnableConfig):
# Convert message dicts to HumanMessage instances if needed.
system_message="""
Summarize the current conversation in no more than
1000 words. Also retain any unanswered quiz questions along with
your internal answers.
"""
messages = []
for m in state["messages"]:
if isinstance(m, dict):
# Use role from dict (defaulting to 'user' if missing)
messages.append(AIMessage(content=system_message, role=m.get("role", "assistant")))
else:
messages.append(m)
summaries = llm_with_tools.invoke(messages)
summary_content = summaries.content
# Call Tool Manually
message_with_single_tool_call = AIMessage(
content="",
tool_calls=[
{
"name": "Summary Tool",
"args": {"summary_text": summary_content},
"id": "tool_call_id",
"type": "tool_call",
}
],
)
tool_node.invoke({"messages": [message_with_single_tool_call]})
###############################################################################
# 5. Build the Graph
###############################################################################
graph_builder = StateGraph(MessagesState)
# Use the built-in ToolNode from langgraph that calls any declared tools.
tools = [
mcq_retriever_tool,
web_extraction_tool,
ensemble_retriever_tool,
general_retriever_tool,
in_memory_retriever_tool,
recall_memory_tool,
summarise_tool,
]
tool_node = ToolNode(tools=tools)
#end_node = ToolNode(tools=[summarise_tool])
# Wrap your model with tools
llm_with_tools = model.bind_tools(tools)
###############################################################################
# 6. The agent's main node: call_model
###############################################################################
def call_model(state: MessagesState, config: RunnableConfig):
"""
The main agent node that calls the LLM with the user + system messages.
Since our vLLM chat wrapper expects a list of BaseMessage objects,
we convert any dict messages to HumanMessage objects.
If the LLM requests a tool call, we'll route to the 'tools' node next
(depending on the condition).
"""
# Convert message dicts to HumanMessage instances if needed.
messages = []
for m in state["messages"]:
if isinstance(m, dict):
# Use role from dict (defaulting to 'user' if missing)
messages.append(HumanMessage(content=m.get("content", ""), role=m.get("role", "user")))
else:
messages.append(m)
# Invoke the LLM (with tools) using the converted messages.
response = llm_with_tools.invoke(messages)
return {"messages": [response]}
def call_summary(state: MessagesState, config: RunnableConfig):
# Convert message dicts to HumanMessage instances if needed.
system_message="""
Summarize the current conversation in no more than
1000 words. Also retain any unanswered quiz questions along with
your internal answers.
"""
messages = []
for m in state["messages"]:
if isinstance(m, dict):
# Use role from dict (defaulting to 'user' if missing)
messages.append(AIMessage(content=system_message, role=m.get("role", "assistant")))
else:
messages.append(m)
summaries = llm_with_tools.invoke(messages)
summary_content = summaries.content
# Call Tool Manually
message_with_single_tool_call = AIMessage(
content="",
tool_calls=[
{
"name": "Summary Tool",
"args": {"summary_text": summary_content},
"id": "tool_call_id",
"type": "tool_call",
}
],
)
tool_node.invoke({"messages": [message_with_single_tool_call]})
###############################################################################
# 7. Add Nodes & Edges, Then Compile
###############################################################################
graph_builder.add_node("agent", call_model)
graph_builder.add_node("tools", tool_node)
#graph_builder.add_node("summary", call_summary)
# Entry point
graph_builder.set_entry_point("agent")
# def custom_tools_condition(llm_output: dict) -> str:
# """Return which node to go to next based on the LLM output."""
# # The LLM's JSON might have a field like {"name": "Recall Memory Tool", "arguments": {...}}.
# tool_name = llm_output.get("name", None)
# # If the LLM calls "Summary Tool", jump directly to the 'summary' node
# if tool_name == "Summary Tool":
# return "summary"
# # If the LLM calls any other recognized tool, go to 'tools'
# valid_tool_names = [t.name for t in tools] # all tools in the main tool_node
# if tool_name in valid_tool_names:
# return "tools"
# # If there's no recognized tool name, assume we're done => go to summary
# return "__end__"
# graph_builder.add_conditional_edges(
# "agent",
# custom_tools_condition,
# {
# "tools": "tools",
# "summary": "summary",
# "__end__": "summary",
# }
# )
# If LLM requests a tool, go to "tools", otherwise go to "summary"
graph_builder.add_conditional_edges("agent", tools_condition)
#graph_builder.add_conditional_edges("agent", tools_condition, {"tools": "tools", "__end__": "summary"})
#graph_builder.add_conditional_edges("agent", lambda llm_output: "tools" if llm_output.get("name", None) in [t.name for t in tools] else "summary", {"tools": "tools", "__end__": "summary"}
# If we used a tool, return to the agent for final answer or more tools
graph_builder.add_edge("tools", "agent")
#graph_builder.add_edge("agent", "summary")
#graph_builder.set_finish_point("summary")
# Compile the graph with checkpointing and persistent store
graph = graph_builder.compile(checkpointer=checkpointer, store=in_memory_store)
#from langgraph.prebuilt import create_react_agent
#graph = create_react_agent(llm_with_tools, tools=tool_node, checkpointer=checkpointer, store=in_memory_store)
#from IPython.display import Image, display
#display(Image(graph.get_graph().draw_mermaid_png()))
########################################
# Gradio Chatbot Application
########################################
import gradio as gr
from gradio import ChatMessage
system_prompt = "You are a helpful Assistant. You will always use the tools available to you from {tools} to address user queries."
########################################
# Upload_documents
########################################
def upload_documents(file_list: List):
"""
Load documents into in-memory vector store.
"""
_documents = []
for doc_path in file_list:
_documents.extend(load_file(doc_path))
# Split the documents into smaller chunks for better embedding coverage
splitter = RecursiveCharacterTextSplitter(
chunk_size=300, # Split into chunks of 512 characters
chunk_overlap=50, # Overlap by 50 characters
add_start_index=True
)
chunked_texts = splitter.split_documents(_documents)
in_memory_vs.add_documents(chunked_texts)
return f"Uploaded {len(file_list)} documents into in-memory vector store."
########################################
# Submit_queries (ChatInterface Function)
########################################
def submit_queries(message, _messages):
"""
- message: dict with {"text": ..., "files": [...]}
- history: list of ChatMessage
"""
_messages=[]
user_text = message.get("text", "")
user_files = message.get("files", [])
# Process user-uploaded files
if user_files:
for file_obj in user_files:
_messages.append(ChatMessage(role="user", content=f"Uploaded file: {file_obj}"))
upload_response = upload_documents(user_files)
_messages.append(ChatMessage(role="assistant", content=upload_response))
yield _messages
return # Exit early since we don't need to process text or call the LLM
# Append user text if present
if user_text:
events = graph.stream(
{
"messages": [
{"role": "system", "content": system_prompt},
{"role": "user", "content": user_text},
]
},
config,
stream_mode="values"
)
for event in events:
response = event["messages"][-1]
if isinstance(response, AIMessage):
if "tool_calls" in response.additional_kwargs:
_messages.append(
ChatMessage(role="assistant",
content=str(response.tool_calls[0]["args"]),
metadata={"title": str(response.tool_calls[0]["name"]),
"id": config["configurable"]["thread_id"]
}
))
yield _messages
else:
_messages.append(ChatMessage(role="assistant",
content=response.content,
metadata={"id": config["configurable"]["thread_id"]
}
))
yield _messages
return _messages
########################################
# 3) Save Chat History
########################################
CHAT_HISTORY_FILE = "chat_history.json"
def save_chat_history(history):
"""
Saves the chat history into a JSON file.
"""
session_history = [
{
"role": "user" if msg.is_user else "assistant",
"content": msg.content
}
for msg in history
]
with open(CHAT_HISTORY_FILE, "w", encoding="utf-8") as f:
json.dump(session_history, f, ensure_ascii=False, indent=4)
########################################
# 6) Main Gradio Interface
########################################
with gr.Blocks() as AI_Tutor:
gr.Markdown("# AI Tutor Chatbot (Gradio App)")
# Primary Chat Interface
chat_interface = gr.ChatInterface(
fn=submit_queries,
type="messages",
chatbot=gr.Chatbot(
label="Chat Window",
height=500,
type="messages"
),
textbox=gr.MultimodalTextbox(
interactive=True,
file_count="multiple",
file_types=[".pdf",".ppt",".pptx",".doc",".docx",".md","image"],
sources=["upload"],
label="Type your query here:",
placeholder="Enter your question...",
),
title="AI Tutor Chatbot",
description="Ask me anything about Artificial Intelligence!",
multimodal=True,
save_history=True,
)
if __name__ == "__main__":
AI_Tutor.launch(server_name="0.0.0.0", server_port=7860)
|