File size: 39,457 Bytes
3efe7a4 |
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 |
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "rkRTHAoXbOJC",
"outputId": "ab776f45-7c6c-4b1c-87bc-7410dc1955fe"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Selecting previously unselected package cloudflared.\n",
"(Reading database ... 126441 files and directories currently installed.)\n",
"Preparing to unpack cloudflared-linux-amd64.deb ...\n",
"Unpacking cloudflared (2025.9.1) ...\n",
"Setting up cloudflared (2025.9.1) ...\n",
"Processing triggers for man-db (2.10.2-1) ...\n",
"cloudflared version 2025.9.1 (built 2025-09-22-13:28 UTC)\n"
]
}
],
"source": [
"!pip install -r requirements.txt -q\n",
"!pip install streamlit cloudflared -q\n",
"!wget -q https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb\n",
"!dpkg -i cloudflared-linux-amd64.deb\n",
"\n",
"!cloudflared --version\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "UpQo5rPBkvT4"
},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "l08lsc3SbUy2",
"outputId": "e7c5db50-4944-4fad-bad6-fae2ec7439aa"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"✅ CUDA is available. Using GPU: Tesla T4\n"
]
}
],
"source": [
"import torch\n",
"\n",
"if torch.cuda.is_available():\n",
" print(f\"✅ CUDA is available. Using GPU: {torch.cuda.get_device_name(0)}\")\n",
" # return True\n",
"else:\n",
" print(\"⚠️ CUDA not available. Falling back to CPU.\")\n",
" # return False\n",
"\n",
"\n",
"# # Load the allocator\n",
"# new_alloc = torch.cuda.memory.CUDAPluggableAllocator(\n",
"# 'alloc.so', 'my_malloc', 'my_free')\n",
"# # Swap the current allocator\n",
"# torch.cuda.memory.change_current_allocator(new_alloc)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "LHHSaPwNbZXW",
"outputId": "a2939de4-7a06-4a35-cf6f-190ea3fec13a"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Overwriting Embeddings.py\n"
]
}
],
"source": [
"%%writefile Embeddings.py\n",
"\n",
"import os\n",
"import glob\n",
"import pickle, json\n",
"from tqdm import tqdm\n",
"import numpy as np\n",
"\n",
"# Try imports with friendly errors\n",
"try:\n",
" import faiss\n",
"except Exception as e:\n",
" raise ImportError(\"faiss is required. Install cpu version: `pip install faiss-cpu` or install via conda for GPU (faiss-gpu).\") from e\n",
"\n",
"try:\n",
" from sentence_transformers import SentenceTransformer\n",
"except Exception as e:\n",
" raise ImportError(\"sentence-transformers is required. `pip install sentence-transformers`\") from e\n",
"\n",
"from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline\n",
"import torch\n",
"from google.colab import userdata\n",
"\n",
"\n",
"\n",
"# from Data_Cleaning import GetDataCleaning\n",
"# from Logger import GetLogger\n",
"\n",
"\n",
"class GetEmbeddings:\n",
" \"\"\"\n",
" Embedding pipeline for cleaned text files.\n",
" Generates embeddings using SentenceTransformers, builds a FAISS index,\n",
" and allows searching queries against the vector database.\n",
" \"\"\"\n",
"\n",
" def __init__(self, config_path=\"config.json\", logger=None):\n",
"\n",
" with open(config_path, \"r\") as f:\n",
" self.config = json.load(f)\n",
"\n",
" cfg_paths = self.config[\"paths\"]\n",
"\n",
"\n",
" cfg_emb = self.config[\"embedding\"]\n",
"\n",
" self.root = cfg_paths[\"root\"]\n",
" self.cleaned_suffix = \"_cleaned_txt\"\n",
" self.chunk_words = cfg_emb[\"chunk_words\"]\n",
" self.batch_size = cfg_emb[\"batch_size\"]\n",
" self.faiss_index_path = cfg_paths[\"faiss_index\"]\n",
" self.metadata_path = cfg_paths[\"metadata\"]\n",
" self.embedding_model = cfg_emb[\"model\"]\n",
"\n",
" # if not logger:\n",
" # obj = GetLogger()\n",
" # logger = obj.get_logger()\n",
" # self.logger = logger\n",
" # print(\"Initializing Embedding Pipeline...\")\n",
"\n",
" # Device\n",
" self.device = \"cuda\" if self.check_cuda() and cfg_emb[\"use_gpu\"] else \"cpu\"\n",
" self.hf_token = \"your_token\"\n",
"\n",
" def check_cuda(self):\n",
" \"\"\"Return True if CUDA is available and usable.\"\"\"\n",
" try:\n",
" if torch.cuda.is_available():\n",
" _ = torch.cuda.current_device()\n",
" print(f\"✅ CUDA available. Device: {torch.cuda.get_device_name(0)}\")\n",
" return True\n",
" print(\"⚠️ CUDA not available. Using CPU.\")\n",
" return False\n",
" except Exception as e:\n",
" print(f\"Error checking CUDA, defaulting to CPU. Error: {e}\")\n",
" return False\n",
"\n",
" def list_cleaned_files(self):\n",
" \"\"\"Return sorted list of cleaned text files under root/*{cleaned_suffix}/*.txt\"\"\"\n",
" pattern = os.path.join(self.root, f\"*{self.cleaned_suffix}\", \"*.txt\")\n",
" files = glob.glob(pattern)\n",
" files.sort()\n",
" return files\n",
"\n",
" def read_text_file(self, path):\n",
" \"\"\"Read a text file and return string content.\"\"\"\n",
" with open(path, \"r\", encoding=\"utf-8\") as f:\n",
" return f.read()\n",
"\n",
" def chunk_text_words(self, text):\n",
" \"\"\"\n",
" Simple word-based chunking.\n",
" Returns list of text chunks.\n",
" \"\"\"\n",
" words = text.split()\n",
" if not words:\n",
" return []\n",
" return [\" \".join(words[i:i + self.chunk_words]) for i in range(0, len(words), self.chunk_words)]\n",
"\n",
" def save_index_and_metadata(self):\n",
" \"\"\"Save FAISS index and metadata to disk.\"\"\"\n",
" os.makedirs(os.path.dirname(self.faiss_index_path), exist_ok=True)\n",
" faiss.write_index(self.index, self.faiss_index_path)\n",
" with open(self.metadata_path, \"wb\") as f:\n",
" pickle.dump(self.metadata, f)\n",
" print(f\"💾 Saved FAISS index to {self.faiss_index_path}\")\n",
" print(f\"💾 Saved metadata to {self.metadata_path}\")\n",
"\n",
" def load_index_and_metadata(self):\n",
" \"\"\"Load FAISS index and metadata if they exist.\"\"\"\n",
" if os.path.exists(self.faiss_index_path) and os.path.exists(self.metadata_path):\n",
" try:\n",
" self.index = faiss.read_index(self.faiss_index_path)\n",
" with open(self.metadata_path, \"rb\") as f:\n",
" self.metadata = pickle.load(f)\n",
" print(f\"✅ Loaded existing FAISS index + metadata from disk.\")\n",
" return True\n",
" except Exception as e:\n",
" print(f\"⚠️ Failed to load FAISS index/metadata, will rebuild. Error: {e}\")\n",
" return False\n",
" return False\n",
"\n",
" def load_encoder(self):\n",
" \"\"\"Loading Encoder\"\"\"\n",
" self.encoder = SentenceTransformer(self.embedding_model, device=self.device)\n",
" print(f\"Loaded embedding model '{self.embedding_model}' on {self.device}\")\n",
" return self.encoder\n",
"\n",
"\n",
" def building_embeddings_index(self, files):\n",
" \"\"\"Build embeddings for all text chunks and return FAISS index + metadata.\"\"\"\n",
"\n",
"\n",
" all_embeddings, metadata = [], []\n",
" next_id = 0\n",
" # Iterate files and chunks\n",
" for fp in tqdm(files, desc=\"Files\", unit=\"file\"):\n",
" text = self.read_text_file(fp)\n",
"\n",
" if not text.strip():\n",
" continue\n",
"\n",
" # metadata: infer company and file from path\n",
" # e.g., financial_reports/Infosys_cleaned_txt/Infosys_2023_AR.txt\n",
" rel = os.path.relpath(fp, self.root)\n",
" folder = rel.split(os.sep)[0]\n",
" filename = os.path.basename(fp)\n",
"\n",
" chunks = self.chunk_text_words(text)\n",
" if not chunks:\n",
" continue\n",
"\n",
" for i in range(0, len(chunks), self.batch_size):\n",
" batch = chunks[i:i + self.batch_size]\n",
" embs = self.encoder.encode(batch, show_progress_bar=False, convert_to_numpy=True)\n",
" embs = embs.astype(np.float32)\n",
"\n",
" for j, vec in enumerate(embs):\n",
" all_embeddings.append(vec)\n",
" metadata.append({\n",
" \"id\": next_id,\n",
" \"source_folder\": folder,\n",
" \"file\": filename,\n",
" \"chunk_id\": i + j,\n",
" \"text\": batch[j] # store chunk text for retrieval\n",
" })\n",
" next_id += 1\n",
"\n",
" if not all_embeddings:\n",
" raise RuntimeError(\"No embeddings were produced. Check cleaned files and chunking.\")\n",
"\n",
" emb_matrix = np.vstack(all_embeddings).astype(np.float32)\n",
" faiss.normalize_L2(emb_matrix)\n",
"\n",
" # Build FAISS index (IndexFlatIP over normalized vectors = cosine similarity)\n",
" dim = emb_matrix.shape[1]\n",
" self.index = faiss.IndexFlatIP(dim)\n",
" self.index.add(emb_matrix)\n",
" self.metadata = metadata\n",
" print(f\"✅ Built FAISS index with {self.index.ntotal} vectors, dim={dim}\")\n",
"\n",
" return self.index, self.metadata\n",
"\n",
" def run(self):\n",
" \"\"\"Main entry: load or build embeddings + FAISS index.\"\"\"\n",
" if self.load_index_and_metadata():\n",
" return\n",
"\n",
" files = self.list_cleaned_files()\n",
" if not files:\n",
" print(\"❌ No cleaned text files found.\")\n",
" raise SystemExit(1)\n",
" self.load_encoder()\n",
" self.building_embeddings_index(files)\n",
" self.save_index_and_metadata()\n",
"\n",
" def load_summarizer(self, model_name=\"google/gemma-2b\"):\n",
" \"\"\"\n",
" Load summarizer LLM once.\n",
" If already loaded, skip.\n",
" \"\"\"\n",
" if hasattr(self, \"summarizer_pipeline\"):\n",
" print(\"ℹ️ Summarizer already loaded, skipping reload.\")\n",
" return\n",
"\n",
" try:\n",
" print(f\"⏳ Loading summarizer model '{model_name}'...\")\n",
" self.tokenizer = AutoTokenizer.from_pretrained(model_name, token=self.hf_token)\n",
" self.summarizer_model = AutoModelForCausalLM.from_pretrained(\n",
" model_name,\n",
" torch_dtype=torch.float16 if self.device == \"cuda\" else torch.float32,\n",
" device_map=self.device,\n",
" token=self.hf_token\n",
" )\n",
" self.summarizer_pipeline = pipeline(\n",
" \"text-generation\",\n",
" model=self.summarizer_model,\n",
" tokenizer=self.tokenizer\n",
" )\n",
" print(f\"✅ Summarizer model '{model_name}' loaded successfully.\")\n",
"\n",
" except RuntimeError as e:\n",
" if \"CUDA out of memory\" in str(e):\n",
" print(\"⚠️ CUDA OOM while loading summarizer. Retrying on CPU...\")\n",
" self.device = \"cpu\"\n",
" torch.cuda.empty_cache()\n",
" return self.load_summarizer(model_name=model_name)\n",
" else:\n",
" print(f\"❌ Failed to load summarizer: {e}\")\n",
" raise\n",
"\n",
" def summarize_chunks(self, chunks, max_content_tokens=2048, max_output_tokens=256):\n",
" \"\"\"\n",
" Summarize list of text chunks using LLM.\n",
" - Chunks are joined until they fit into max_context_tokens\n",
" - Generates a concise summary.\n",
" \"\"\"\n",
"\n",
" if not hasattr(self, \"summarizer_pipeline\"):\n",
" self.load_summarizer()\n",
" print(\"Summarizer not initialized. Called load_summarizer(). pipeline will work with default parameters.\")\n",
"\n",
" # Join chunks into one context, respecting token budget\n",
" context = \" \".join(chunks)\n",
" input_tokens = len(self.tokenizer.encode(context))\n",
"\n",
" if input_tokens > max_content_tokens:\n",
" # Trim to fit context window\n",
" context = \" \".join(context.split()[:max_content_tokens])\n",
" print(\"⚠️ Context truncated to fit within model token limit.\")\n",
"\n",
" # Build summarization prompt\n",
" prompt = f\"\"\"\n",
" Summarize the following financial report excerpts into a concise answer.\n",
" Keep it factual, short, and grounded in the text.\n",
"\n",
" Excerpts:\n",
" {context}\n",
"\n",
" Summary:\n",
" \"\"\"\n",
"\n",
" try:\n",
" output = self.summarizer_pipeline(\n",
" prompt,\n",
" max_new_tokens=max_output_tokens,\n",
" do_sample=False\n",
" )[0][\"generated_text\"]\n",
"\n",
" if \"Summary:\" in output:\n",
" summary = output.split(\"Summary:\")[-1].strip()\n",
" else:\n",
" summary = output.strip()\n",
"\n",
" return summary\n",
"\n",
" except RuntimeError as e:\n",
" if \"CUDA out of memory\" in str(e):\n",
" print(\"⚠️ CUDA OOM during summarization. Retrying on CPU...\")\n",
" self.device = \"cpu\"\n",
" torch.cuda.empty_cache()\n",
" return self.summarize_chunks(chunks, max_content_tokens, max_output_tokens)\n",
" else:\n",
" print(f\"❌ Summarizer failed: {e}. Falling back to raw chunks.\")\n",
" return \" \".join(chunks[:2]) # fallback: return first 2 chunks\n",
"\n",
"\n",
" def answer_query(self, query, top_k=3):\n",
" \"\"\"\n",
" End-to-end QA:\n",
" - Retrieve relevant chunks from FAISS\n",
" - Summarize into a final answer.\n",
" \"\"\"\n",
" try:\n",
" #step 1: Retrieve\n",
" print(f\"🔍 searching vector DB for query: {query}\")\n",
" q_emb = self.encoder.encode(query, show_progress_bar=False, convert_to_numpy=True).reshape(1, -1)\n",
" faiss.normalize_L2(q_emb)\n",
"\n",
" scores, idxs = self.index.search(q_emb, k=top_k)\n",
" chunks = [self.metadata[idx][\"text\"] for idx in idxs[0]]\n",
"\n",
" # Step 2: Summarize\n",
" summary = self.summarize_chunks(chunks)\n",
"\n",
" # Log results\n",
" print(f\"✅ Final Answer: {summary}\")\n",
" return summary\n",
"\n",
" except Exception as e:\n",
" print(f\"Error in answer_query: {e}\")\n",
" return None\n",
"\n",
"\n",
"# Example\n",
"# ge = GetEmbeddings()\n",
"# ge.run()\n",
"# # NEW STEP\n",
"# ge.load_summarizer(\"google/gemma-2b\")\n",
"# answer = ge.answer_query(\"What are the key highlights from Q2 financial report?\")\n",
"# print(answer)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "SrZwOeGPba8Q",
"outputId": "b14f3d67-54d7-4db1-c030-702ab670bc90"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Writing Evaluator.py\n"
]
}
],
"source": [
"%%writefile Evaluator.py\n",
"import os\n",
"import json\n",
"import time\n",
"import numpy as np\n",
"from tqdm import tqdm\n",
"\n",
"# from Logger import GetLogger, MetricsLogger\n",
"# from Embeddings import GetEmbeddings\n",
"\n",
"# Metrics\n",
"from sklearn.metrics.pairwise import cosine_similarity\n",
"from rouge_score import rouge_scorer\n",
"from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction\n",
"from bert_score import score as bert_score\n",
"\n",
"class Evaluator:\n",
" \"\"\"\n",
" Evaluation pipeline for financial QA Agent.\n",
" Uses eval_dataset.json to run queries, collect answers, and compute metrics.\n",
" \"\"\"\n",
" def __init__(self, config_path=\"config.json\", logger=None):\n",
" with open(config_path, \"r\") as f:\n",
" self.config = json.load(f)\n",
" self.paths = self.config[\"paths\"]\n",
"\n",
"\n",
" # if not logger:\n",
" # obj = GetLogger()\n",
" # logger = obj.get_logger()\n",
" # self.logger = logger\n",
"\n",
"\t\t# # Metrics logger\n",
" # self.metrics_logger = MetricsLogger(logger=self.logger)\n",
"\n",
" # Initialize Agent\n",
" self.agent = GetEmbeddings(config_path=config_path, logger=None)\n",
" self.agent.run() # Load or rebuild FAISS + embeddings\n",
" self.agent.load_summarizer() # Load summarizer\n",
" self.encoder = self.agent.load_encoder()\n",
"\n",
" # Load Dataset\n",
" self.dataset = self.load_dataset()\n",
" self.results = []\n",
" self.failed_queries = []\n",
"\n",
" def load_dataset(self):\n",
" path = self.paths[\"eval_dataset\"]\n",
" if not os.path.exists(path):\n",
" raise FileNotFoundError(f\"Dataset not found: {path}\")\n",
" with open(path, \"r\", encoding=\"utf-8\") as f:\n",
" return json.load(f)\n",
"\n",
" def measure_latency(self, func, *args, **kwargs):\n",
" \"\"\"Helper: measure time taken by a function call.\"\"\"\n",
" start = time.time()\n",
" result = func(*args, **kwargs)\n",
" latency = time.time() - start\n",
" return result, latency\n",
"\n",
" def evaluate_query(self, query, reference):\n",
" \"\"\"Run one query, compare answer vs. reference, compute metrics.\"\"\"\n",
" # try:\n",
" # Run pipeline\n",
" system_answer, latency = self.measure_latency(self.agent.answer_query, query)\n",
"\n",
" # 1. Embedding similarity (proxy retrieval quality)\n",
" ref_emb = self.encoder.encode([reference], convert_to_numpy=True)\n",
" ans_emb = self.encoder.encode([system_answer], convert_to_numpy=True)\n",
" retrieval_quality = float(cosine_similarity(ref_emb, ans_emb)[0][0])\n",
"\n",
" # 2. ROUGE-L\n",
" scorer = rouge_scorer.RougeScorer(['rougeL'], use_stemmer=True)\n",
" rouge_score = scorer.score(reference, system_answer)['rougeL'].fmeasure\n",
"\n",
" # 3. BLEU (with smoothing for short texts)\n",
" smoothie = SmoothingFunction().method4\n",
" bleu = sentence_bleu([reference.split()], system_answer.split(), smoothing_function=smoothie)\n",
"\n",
" # 4. BERTScore (semantic similarity)\n",
" P, R, F1 = bert_score([system_answer], [reference], lang=\"en\")\n",
" bert_f1 = float(F1.mean())\n",
"\n",
" metrics = {\n",
" \"query\": query,\n",
" \"reference\": reference,\n",
" \"system_answer\": system_answer,\n",
" \"retrieval_quality\": retrieval_quality,\n",
" \"rougeL\": rouge_score,\n",
" \"bleu\": bleu,\n",
" \"bertscore_f1\": bert_f1,\n",
" \"latency_sec\": latency\n",
" }\n",
"\n",
" # Log into metrics logger\n",
" # self.metrics_logger.log_query_metrics(query, metrics)\n",
"\n",
" return metrics\n",
"\n",
" # except Exception as e:\n",
" # print(f\"Error evaluating query '{query}': {e}\")\n",
" # return None\n",
"\n",
"\n",
" def run(self):\n",
" \"\"\"Run evaluation on entire dataset.\"\"\"\n",
" print(\"Starting Evaluation...\")\n",
"\n",
" for item in tqdm(self.dataset, desc=\"Queries\"):\n",
" query = item[\"query\"]\n",
" reference = item[\"reference\"]\n",
" result = self.evaluate_query(query, reference)\n",
" if result:\n",
" self.results.append(result)\n",
"\n",
"\n",
" # Save result\n",
" with open(self.paths[\"eval_results\"], \"w\", encoding=\"utf-8\") as f:\n",
" json.dump(self.results, f, indent=2)\n",
"\n",
" if self.failed_queries:\n",
" with open(self.paths[\"failed_queries\"], \"w\", encoding=\"utf-8\") as f:\n",
" json.dump(self.failed_queries, f, indent=2)\n",
"\n",
"\n",
" # Save metrics summary\n",
" # summary = self.metrics_logger.save()\n",
" summary = None\n",
" print(f\"Evaluation Complete.\")\n",
" print(f\"📊 Evaluation summary: {summary}\")\n",
"\n",
" return self.results, summary\n",
"\n",
"\n",
"if __name__ == \"__main__\":\n",
" evaluator = Evaluator()\n",
" results, summary = evaluator.run()\n",
"\n",
" print(\"\\n=== Sample Results ===\")\n",
" print(json.dumps(results[:2], indent=2))\n",
" print(\"\\n=== Summary ===\")\n",
" print(json.dumps(summary, indent=2))\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "_SgMUhSJbdcu",
"outputId": "c79fe42b-517f-40b7-cc2b-71ddaae05084"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Overwriting app.py\n"
]
}
],
"source": [
"%%writefile app.py\n",
"import streamlit as st\n",
"import json\n",
"import faiss\n",
"import numpy as np\n",
"import re\n",
"from Embeddings import GetEmbeddings\n",
"from Logger import GetLogger\n",
"\n",
"# ================================\n",
"# Load Config\n",
"# ================================\n",
"with open(\"config.json\", \"r\") as f:\n",
" config = json.load(f)\n",
"\n",
"# Initialize Logger\n",
"log_obj = GetLogger()\n",
"logger = log_obj.get_logger()\n",
"\n",
"# Initialize QA Agent\n",
"@st.cache_resource\n",
"def load_agent():\n",
" agent = GetEmbeddings(config_path=\"config.json\", logger=logger)\n",
" agent.run() # load or build FAISS index\n",
" encoder = agent.load_encoder()\n",
" agent.load_summarizer()\n",
" return agent, encoder\n",
"\n",
"agent, encoder = load_agent()\n",
"\n",
"# ================================\n",
"# Streamlit UI\n",
"# ================================\n",
"st.set_page_config(page_title=\"Financial QA Agent\", layout=\"wide\")\n",
"\n",
"# --- Header ---\n",
"st.title(\"💹 Financial Report QA Agent\")\n",
"st.markdown(\n",
" \"\"\"\n",
" Welcome!\n",
" This tool lets you **query annual financial reports** (Infosys, ICICI Bank, etc.)\n",
" and get **summarized answers** with supporting evidence from the text.\n",
" \"\"\"\n",
")\n",
"\n",
"# Sidebar - Settings\n",
"st.sidebar.header(\"⚙️ Settings\")\n",
"top_k = st.sidebar.slider(\"Top K Chunks\", 1, 10, 3)\n",
"max_output_tokens = st.sidebar.slider(\"Max Summary Tokens\", 64, 512, 256)\n",
"\n",
"# --- Keyword highlighting ---\n",
"def highlight_keywords(text, keywords=[\"risk\", \"revenue\", \"profit\", \"growth\", \"loss\"]):\n",
" pattern = re.compile(r\"\\b(\" + \"|\".join(keywords) + r\")\\b\", re.IGNORECASE)\n",
" return pattern.sub(lambda m: f\"**{m.group(0)}**\", text)\n",
"\n",
"# --- Session State for Query History ---\n",
"if \"history\" not in st.session_state:\n",
" st.session_state[\"history\"] = []\n",
"\n",
"# --- Query input ---\n",
"query = st.text_input(\"🔍 Enter your question:\", placeholder=\"e.g., What are the main risk factors in 2023?\")\n",
"\n",
"if st.button(\"Get Answer\"):\n",
" if query.strip() == \"\":\n",
" st.warning(\"Please enter a query.\")\n",
" else:\n",
" with st.spinner(\"Searching reports...\"):\n",
" try:\n",
" # Retrieve + summarize\n",
" answer = agent.answer_query(query, top_k=top_k)\n",
"\n",
" # --- Display final answer ---\n",
" st.subheader(\"📌 Answer\")\n",
" st.success(answer)\n",
"\n",
" # --- Show supporting chunks ---\n",
" st.subheader(\"📂 Supporting Chunks\")\n",
" q_emb = encoder.encode(query, convert_to_numpy=True).reshape(1, -1)\n",
" faiss.normalize_L2(q_emb)\n",
" scores, idxs = agent.index.search(q_emb.astype(np.float32), k=top_k)\n",
"\n",
" for score, idx in zip(scores[0], idxs[0]):\n",
" meta = agent.metadata[idx]\n",
" with st.expander(f\"📄 {meta['file']} | Chunk {meta['chunk_id']} | Score: {score:.4f}\"):\n",
" chunk_text = highlight_keywords(meta['text'][:1000])\n",
" st.markdown(chunk_text)\n",
"\n",
" # --- Save Query & Answer to History ---\n",
" st.session_state[\"history\"].append({\"query\": query, \"answer\": answer})\n",
"\n",
" # --- Log query + answer ---\n",
" logger.info(f\"User Query: {query}\")\n",
" logger.info(f\"System Answer: {answer}\")\n",
"\n",
" # --- Save persistent history JSON ---\n",
" with open(\"ui_query_history.json\", \"w\", encoding=\"utf-8\") as f:\n",
" json.dump(st.session_state[\"history\"], f, indent=2)\n",
"\n",
" except Exception as e:\n",
" st.error(f\"Error: {e}\")\n",
" logger.error(f\"Streamlit UI error: {e}\")\n",
"\n",
"# --- Show History in Sidebar ---\n",
"if st.session_state[\"history\"]:\n",
" st.sidebar.subheader(\"🕘 Query History\")\n",
" for item in st.session_state[\"history\"][-5:]: # show last 5 queries\n",
" st.sidebar.write(f\"**Q:** {item['query']}\")\n",
" st.sidebar.write(f\"**A:** {item['answer'][:100]}...\")\n",
" st.sidebar.markdown(\"---\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "6UAnlclVckzM",
"outputId": "bb65eead-5953-4a4f-f838-14fadc1469dd"
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[90m2025-09-29T13:35:21Z\u001b[0m \u001b[32mINF\u001b[0m Thank you for trying Cloudflare Tunnel. Doing so, without a Cloudflare account, is a quick way to experiment and try it out. However, be aware that these account-less Tunnels have no uptime guarantee, are subject to the Cloudflare Online Services Terms of Use (https://www.cloudflare.com/website-terms/), and Cloudflare reserves the right to investigate your use of Tunnels for violations of such terms. If you intend to use Tunnels in production you should use a pre-created named tunnel by following: https://developers.cloudflare.com/cloudflare-one/connections/connect-apps\n",
"\u001b[90m2025-09-29T13:35:21Z\u001b[0m \u001b[32mINF\u001b[0m Requesting new quick Tunnel on trycloudflare.com...\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m +--------------------------------------------------------------------------------------------+\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m | Your quick Tunnel has been created! Visit it at (it may take some time to be reachable): |\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m | https://ease-library-cases-gibraltar.trycloudflare.com |\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m +--------------------------------------------------------------------------------------------+\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Cannot determine default configuration path. No file [config.yml config.yaml] in [~/.cloudflared ~/.cloudflare-warp ~/cloudflare-warp /etc/cloudflared /usr/local/etc/cloudflared]\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Version 2025.9.1 (Checksum 3dc1dc4252eae3c691861f926e2b8640063a2ce534b07b7a3f4ec2de439ecfe3)\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m GOOS: linux, GOVersion: go1.24.4, GoArch: amd64\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Settings: map[ha-connections:1 no-autoupdate:true protocol:quic url:http://localhost:8501]\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m cloudflared will not automatically update if installed by a package manager.\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Generated Connector ID: b7e0104f-71af-4b1e-a366-b3b15b2c86d9\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Initial protocol quic\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m ICMP proxy will use 172.28.0.12 as source for IPv4\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m ICMP proxy will use :: as source for IPv6\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[1m\u001b[31mERR\u001b[0m\u001b[0m Cannot determine default origin certificate path. No file cert.pem in [~/.cloudflared ~/.cloudflare-warp ~/cloudflare-warp /etc/cloudflared /usr/local/etc/cloudflared]. You need to specify the origin certificate path by specifying the origincert option in the configuration file, or set TUNNEL_ORIGIN_CERT environment variable \u001b[36moriginCertPath=\u001b[0m\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m ICMP proxy will use 172.28.0.12 as source for IPv4\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m ICMP proxy will use :: as source for IPv6\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Starting metrics server on 127.0.0.1:20241/metrics\n",
"\u001b[90m2025-09-29T13:35:25Z\u001b[0m \u001b[32mINF\u001b[0m Tunnel connection curve preferences: [X25519MLKEM768 CurveP256] \u001b[36mconnIndex=\u001b[0m0 \u001b[36mevent=\u001b[0m0 \u001b[36mip=\u001b[0m198.41.200.113\n",
"2025/09/29 13:35:25 failed to sufficiently increase receive buffer size (was: 208 kiB, wanted: 7168 kiB, got: 416 kiB). See https://github.com/quic-go/quic-go/wiki/UDP-Buffer-Sizes for details.\n",
"\u001b[90m2025-09-29T13:35:26Z\u001b[0m \u001b[32mINF\u001b[0m Registered tunnel connection \u001b[36mconnIndex=\u001b[0m0 \u001b[36mconnection=\u001b[0mc535a197-93c0-4941-a9ab-b32533b50549 \u001b[36mevent=\u001b[0m0 \u001b[36mip=\u001b[0m198.41.200.113 \u001b[36mlocation=\u001b[0msin02 \u001b[36mprotocol=\u001b[0mquic\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[32mINF\u001b[0m Initiating graceful shutdown due to signal interrupt ...\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[1m\u001b[31mERR\u001b[0m\u001b[0m failed to run the datagram handler \u001b[31merror=\u001b[0m\u001b[31m\"context canceled\"\u001b[0m \u001b[36mconnIndex=\u001b[0m0 \u001b[36mevent=\u001b[0m0 \u001b[36mip=\u001b[0m198.41.200.113\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[1m\u001b[31mERR\u001b[0m\u001b[0m failed to serve tunnel connection \u001b[31merror=\u001b[0m\u001b[31m\"accept stream listener encountered a failure while serving\"\u001b[0m \u001b[36mconnIndex=\u001b[0m0 \u001b[36mevent=\u001b[0m0 \u001b[36mip=\u001b[0m198.41.200.113\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[1m\u001b[31mERR\u001b[0m\u001b[0m Serve tunnel error \u001b[31merror=\u001b[0m\u001b[31m\"accept stream listener encountered a failure while serving\"\u001b[0m \u001b[36mconnIndex=\u001b[0m0 \u001b[36mevent=\u001b[0m0 \u001b[36mip=\u001b[0m198.41.200.113\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[32mINF\u001b[0m Retrying connection in up to 1s \u001b[36mconnIndex=\u001b[0m0 \u001b[36mevent=\u001b[0m0 \u001b[36mip=\u001b[0m198.41.200.113\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[1m\u001b[31mERR\u001b[0m\u001b[0m Connection terminated \u001b[36mconnIndex=\u001b[0m0\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[1m\u001b[31mERR\u001b[0m\u001b[0m no more connections active and exiting\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[32mINF\u001b[0m Tunnel server stopped\n",
"\u001b[90m2025-09-29T13:38:58Z\u001b[0m \u001b[32mINF\u001b[0m Metrics server stopped\n"
]
}
],
"source": [
"import threading, os\n",
"\n",
"# Kill anything on port 8501 (just in case)\n",
"os.system(\"kill -9 $(lsof -t -i:8501) 2>/dev/null\")\n",
"\n",
"# Run Streamlit in background\n",
"def run_app():\n",
" os.system(\"streamlit run app.py --server.port 8501\")\n",
"\n",
"thread = threading.Thread(target=run_app)\n",
"thread.start()\n",
"\n",
"# Start cloudflared tunnel\n",
"!cloudflared tunnel --url http://localhost:8501 --no-autoupdate\n"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"gpuType": "T4",
"machine_shape": "hm",
"provenance": []
},
"kernelspec": {
"display_name": "Python 3",
"name": "python3"
},
"language_info": {
"name": "python"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
|