File size: 24,384 Bytes
3326079 |
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 |
#!/usr/bin/env python3
"""Embedding generation CLI script for the RAG pipeline.
This script generates embeddings from text chunks and builds search indexes
for the pythermalcomfort RAG chatbot. It supports:
- Reading chunks from JSONL files
- Generating embeddings using BGE encoder with GPU acceleration
- Building FAISS indexes for dense retrieval
- Building BM25 indexes for sparse retrieval
- Publishing artifacts to HuggingFace (optional)
Usage:
# Basic embedding generation
poetry run python scripts/embed.py data/chunks/chunks.jsonl data/embeddings/
# With HuggingFace publishing
poetry run python scripts/embed.py data/chunks/chunks.jsonl data/embeddings/ \
--publish
# Custom batch size and model
poetry run python scripts/embed.py data/chunks/chunks.jsonl data/embeddings/ \
--batch-size 64 --model BAAI/bge-base-en-v1.5
Output Files:
{output_dir}/
βββ embeddings.parquet # Embeddings with chunk_id mapping
βββ metadata.json # Model metadata
βββ faiss_index.bin # FAISS index for dense retrieval
βββ faiss_index.bin.ids.json # Chunk ID mapping for FAISS
βββ bm25_index.pkl # BM25 index for sparse retrieval
βββ chunks.parquet # Chunks in parquet format (if --publish)
"""
from __future__ import annotations
import argparse
import json
import logging
import sys
import time
from pathlib import Path
from typing import TYPE_CHECKING
# =============================================================================
# Environment Variable Loading
# =============================================================================
# Load environment variables from .env file at script startup.
# This ensures HF_TOKEN and other secrets are available before they're needed.
# The .env file should be in the project root directory.
# =============================================================================
from dotenv import load_dotenv
# Find the project root (parent of scripts/ directory) and load .env from there
_PROJECT_ROOT = Path(__file__).parent.parent
_ENV_FILE = _PROJECT_ROOT / ".env"
if _ENV_FILE.exists():
load_dotenv(_ENV_FILE)
# Rich is lightweight - import at module level for progress display
from rich.console import Console
from rich.progress import (
BarColumn,
MofNCompleteColumn,
Progress,
SpinnerColumn,
TaskID,
TaskProgressColumn,
TextColumn,
TimeElapsedColumn,
TimeRemainingColumn,
)
from rich.table import Table
if TYPE_CHECKING:
from rag_chatbot.chunking.models import Chunk
from rag_chatbot.embeddings import (
BGEEncoder,
EmbeddingRecord,
)
# =============================================================================
# Module Constants
# =============================================================================
# Default embedding model (BAAI General Embedding - small variant)
DEFAULT_MODEL: str = "BAAI/bge-small-en-v1.5"
# Default batch size for embedding generation
DEFAULT_BATCH_SIZE: int = 32
# Default embedding dimension for bge-small-en-v1.5
BGE_SMALL_DIM: int = 384
# =============================================================================
# Logging Configuration
# =============================================================================
# Configure logging to stderr with timestamp
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(levelname)s - %(message)s",
handlers=[logging.StreamHandler(sys.stderr)],
)
logger = logging.getLogger(__name__)
# Rich console for styled output
console = Console()
# =============================================================================
# Helper Functions
# =============================================================================
def get_device_info() -> tuple[str, str]:
"""Get GPU/device information for reporting.
This function checks for CUDA availability and returns device information
for display in the CLI output. It imports torch lazily to avoid loading
heavy dependencies until needed.
Returns:
-------
Tuple of (device_type, device_name) where:
- device_type: "cuda" or "cpu"
- device_name: GPU name (e.g., "NVIDIA RTX 4090") or "CPU"
Example:
-------
>>> device_type, device_name = get_device_info()
>>> print(f"Using {device_name}")
Using NVIDIA RTX 4090
"""
# Lazy import torch to avoid loading until needed
import torch # type: ignore[import-not-found]
if torch.cuda.is_available():
device_type = "cuda"
device_name = torch.cuda.get_device_name(0)
else:
device_type = "cpu"
device_name = "CPU"
return device_type, device_name
def load_chunks_from_jsonl(input_path: Path) -> list[Chunk]:
"""Load chunks from a JSONL file.
Reads a JSONL file containing chunk data and parses each line into
a Chunk model instance. Text normalization is applied during loading
to fix common PDF extraction artifacts.
Args:
----
input_path: Path to the chunks.jsonl file.
Returns:
-------
List of Chunk objects parsed from the file.
Raises:
------
FileNotFoundError: If the input file doesn't exist.
ValueError: If the file is empty or contains invalid JSON.
Example:
-------
>>> chunks = load_chunks_from_jsonl(Path("data/chunks/chunks.jsonl"))
>>> len(chunks)
1500
"""
# Lazy import to avoid loading heavy dependencies at module level
from rag_chatbot.chunking.models import Chunk, TextNormalizer
if not input_path.exists():
msg = f"Input file not found: {input_path}"
raise FileNotFoundError(msg)
chunks: list[Chunk] = []
normalizer = TextNormalizer()
with open(input_path, encoding="utf-8") as f:
for line_num, raw_line in enumerate(f, start=1):
# Skip empty lines
line = raw_line.strip()
if not line:
continue
try:
# Parse JSON line
data = json.loads(line)
# Normalize text before creating chunk
# Apply text normalization to fix OCR artifacts
if "text" in data:
data["text"] = normalizer.normalize(data["text"], is_heading=False)
# Create Chunk instance
chunk = Chunk(**data)
chunks.append(chunk)
except json.JSONDecodeError as exc:
logger.warning("Invalid JSON at line %d: %s", line_num, exc)
continue
except Exception as exc:
logger.warning("Error parsing chunk at line %d: %s", line_num, exc)
continue
if not chunks:
msg = f"No valid chunks found in {input_path}"
raise ValueError(msg)
return chunks
def create_embedding_records(
chunks: list[Chunk],
encoder: BGEEncoder,
batch_size: int,
progress: Progress,
task_id: TaskID,
) -> list[EmbeddingRecord]:
"""Generate embeddings for chunks and create EmbeddingRecord objects.
This function encodes all chunk texts using the BGE encoder and creates
EmbeddingRecord instances with chunk_id, chunk_hash, and embedding data.
Progress is reported through the Rich progress bar.
Args:
----
chunks: List of Chunk objects to embed.
encoder: BGEEncoder instance for generating embeddings.
batch_size: Number of chunks to process per batch.
progress: Rich Progress instance for progress tracking.
task_id: Task ID for the progress bar.
Returns:
-------
List of EmbeddingRecord objects with generated embeddings.
Example:
-------
>>> records = create_embedding_records(chunks, encoder, 32, progress, task_id)
>>> len(records) == len(chunks)
True
"""
# Lazy import
from rag_chatbot.embeddings import EmbeddingRecord
# Extract texts for embedding
texts = [chunk.text for chunk in chunks]
# Track progress through callback
def progress_callback(current_batch: int, _total_batches: int) -> None:
"""Update progress bar after each batch."""
progress.update(task_id, completed=current_batch)
# Generate embeddings with progress tracking
embeddings = encoder.encode(
texts=texts,
batch_size=batch_size,
show_progress=False, # We use our own progress bar
progress_callback=progress_callback,
)
# Create EmbeddingRecord for each chunk
records: list[EmbeddingRecord] = []
for idx, chunk in enumerate(chunks):
record = EmbeddingRecord(
chunk_id=chunk.chunk_id,
chunk_hash=chunk.chunk_hash,
embedding=embeddings[idx].tolist(),
)
records.append(record)
return records
def build_indexes(
output_dir: Path,
chunks: list[Chunk],
progress: Progress,
) -> tuple[float, float]:
"""Build FAISS and BM25 indexes from embeddings and chunks.
This function builds both dense (FAISS) and sparse (BM25) indexes
for hybrid retrieval. The FAISS index is built from the embeddings
parquet file, while BM25 is built from chunk texts.
Args:
----
output_dir: Directory containing embeddings.parquet and for saving indexes.
chunks: List of Chunk objects for BM25 indexing.
progress: Rich Progress instance for progress tracking.
Returns:
-------
Tuple of (faiss_build_time, bm25_build_time) in seconds.
Example:
-------
>>> faiss_time, bm25_time = build_indexes(output_dir, chunks, progress)
>>> print(f"FAISS: {faiss_time:.2f}s, BM25: {bm25_time:.2f}s")
"""
# Lazy imports
from rag_chatbot.embeddings import BM25IndexBuilder, FAISSIndexBuilder
embeddings_path = output_dir / "embeddings.parquet"
# Build FAISS index
faiss_task = progress.add_task("[cyan]Building FAISS index...", total=1)
faiss_start = time.perf_counter()
faiss_builder = FAISSIndexBuilder()
faiss_index = faiss_builder.build_from_parquet(embeddings_path)
faiss_builder.save_index(faiss_index, output_dir / "faiss_index.bin")
faiss_time = time.perf_counter() - faiss_start
progress.update(faiss_task, completed=1)
# Build BM25 index
bm25_task = progress.add_task("[cyan]Building BM25 index...", total=1)
bm25_start = time.perf_counter()
bm25_builder = BM25IndexBuilder()
bm25_index, chunk_ids = bm25_builder.build_from_chunks(chunks)
bm25_builder.save_index(bm25_index, chunk_ids, output_dir / "bm25_index.pkl")
bm25_time = time.perf_counter() - bm25_start
progress.update(bm25_task, completed=1)
return faiss_time, bm25_time
def publish_to_huggingface(
output_dir: Path,
chunks: list[Chunk],
model_name: str,
embedding_dim: int,
progress: Progress,
) -> str:
"""Publish all artifacts to HuggingFace dataset repository.
This function handles the complete publishing workflow:
1. Saves chunks to parquet format
2. Generates source manifest
3. Authenticates with HuggingFace
4. Uploads all artifacts
Args:
----
output_dir: Directory containing artifacts to publish.
chunks: List of Chunk objects for chunks.parquet.
model_name: Name of the embedding model used.
embedding_dim: Dimension of embeddings.
progress: Rich Progress instance for progress tracking.
Returns:
-------
URL of the published HuggingFace dataset.
Raises:
------
ValueError: If HF_TOKEN is not set.
RuntimeError: If publishing fails.
Example:
-------
>>> url = publish_to_huggingface(
... output_dir, chunks, "BAAI/bge-small-en-v1.5", 384, progress
... )
>>> print(url)
'https://huggingface.co/datasets/sadickam/pytherm_index'
"""
# Lazy imports
from rag_chatbot.embeddings import HuggingFacePublisher, PublisherConfig
publish_task = progress.add_task("[cyan]Publishing to HuggingFace...", total=4)
# Step 1: Save chunks to parquet
config = PublisherConfig()
publisher = HuggingFacePublisher(config)
publisher.save_chunks_parquet(chunks, output_dir)
progress.update(publish_task, advance=1)
# Step 2: Generate source manifest
manifest = publisher.generate_source_manifest(
source_files=[], # Source files not tracked in this context
total_chunks=len(chunks),
total_embeddings=len(chunks),
)
progress.update(publish_task, advance=1)
# Step 3: Authenticate
publisher.authenticate()
progress.update(publish_task, advance=1)
# Step 4: Publish all artifacts
dataset_url = publisher.publish(
artifacts_dir=output_dir,
manifest=manifest,
model_name=model_name,
embedding_dimension=embedding_dim,
)
progress.update(publish_task, advance=1)
return dataset_url
def print_statistics( # noqa: PLR0913
total_chunks: int,
total_time: float,
embedding_time: float,
faiss_time: float,
bm25_time: float,
device_name: str,
model_name: str,
output_dir: Path,
dataset_url: str | None = None,
) -> None:
"""Print final statistics table using Rich.
Displays a formatted table with embedding statistics including:
- Total chunks processed
- Time breakdowns (embedding, indexing)
- Throughput metrics
- Device information
- Output file sizes
Args:
----
total_chunks: Number of chunks embedded.
total_time: Total elapsed time in seconds.
embedding_time: Time spent on embedding generation.
faiss_time: Time spent building FAISS index.
bm25_time: Time spent building BM25 index.
device_name: Name of the device used (GPU name or "CPU").
model_name: Name of the embedding model.
output_dir: Directory where outputs were saved.
dataset_url: Optional URL of published HuggingFace dataset.
"""
# Calculate throughput
throughput = total_chunks / embedding_time if embedding_time > 0 else 0
# Create statistics table
table = Table(
title="Embedding Statistics", show_header=True, header_style="bold cyan"
)
table.add_column("Metric", style="dim", width=25)
table.add_column("Value", justify="right")
# Add rows
table.add_row("Total Chunks", f"{total_chunks:,}")
table.add_row("Model", model_name)
table.add_row("Device", device_name)
table.add_row("", "") # Separator
table.add_row("Embedding Time", f"{embedding_time:.2f}s")
table.add_row("FAISS Build Time", f"{faiss_time:.2f}s")
table.add_row("BM25 Build Time", f"{bm25_time:.2f}s")
table.add_row("Total Time", f"{total_time:.2f}s")
table.add_row("", "") # Separator
table.add_row("Throughput", f"{throughput:.1f} chunks/sec")
# Add file sizes
embeddings_file = output_dir / "embeddings.parquet"
faiss_file = output_dir / "faiss_index.bin"
bm25_file = output_dir / "bm25_index.pkl"
if embeddings_file.exists():
size_mb = embeddings_file.stat().st_size / (1024 * 1024)
table.add_row("Embeddings Size", f"{size_mb:.2f} MB")
if faiss_file.exists():
size_mb = faiss_file.stat().st_size / (1024 * 1024)
table.add_row("FAISS Index Size", f"{size_mb:.2f} MB")
if bm25_file.exists():
size_mb = bm25_file.stat().st_size / (1024 * 1024)
table.add_row("BM25 Index Size", f"{size_mb:.2f} MB")
# Add dataset URL if published
if dataset_url:
table.add_row("", "") # Separator
table.add_row("Published URL", dataset_url)
console.print()
console.print(table)
console.print()
def parse_args() -> argparse.Namespace:
"""Parse command line arguments.
Returns
-------
Parsed argument namespace with input_path, output_dir, and options.
"""
parser = argparse.ArgumentParser(
description="Generate embeddings and build indexes for RAG retrieval.",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
poetry run python scripts/embed.py chunks.jsonl output/
poetry run python scripts/embed.py chunks.jsonl output/ --publish
poetry run python scripts/embed.py chunks.jsonl output/ --batch-size 64
""",
)
parser.add_argument(
"input_path",
type=Path,
help="Path to chunks.jsonl file containing text chunks",
)
parser.add_argument(
"output_dir",
type=Path,
help="Directory to save embeddings and indexes",
)
parser.add_argument(
"--publish",
action="store_true",
help="Publish artifacts to HuggingFace after embedding",
)
parser.add_argument(
"--batch-size",
type=int,
default=DEFAULT_BATCH_SIZE,
help=f"Batch size for embedding generation (default: {DEFAULT_BATCH_SIZE})",
)
parser.add_argument(
"--model",
type=str,
default=DEFAULT_MODEL,
help=f"Embedding model name (default: {DEFAULT_MODEL})",
)
parser.add_argument(
"--device",
type=str,
default="auto",
choices=["cpu", "cuda", "auto"],
help="Device to use for embedding (default: auto)",
)
return parser.parse_args()
def main() -> int:
"""Run the embedding generation pipeline.
This is the main entry point for the embed.py CLI script. It orchestrates:
1. Loading chunks from JSONL
2. Initializing the encoder with GPU if available
3. Generating embeddings with progress tracking
4. Saving embeddings to parquet storage
5. Building FAISS and BM25 indexes
6. Optionally publishing to HuggingFace
Returns
-------
Exit code (0 for success, 1 for error).
"""
# Parse command line arguments
args = parse_args()
# Track total time
start_time = time.perf_counter()
try:
# =================================================================
# Step 1: Display GPU/Device Information
# =================================================================
console.print("\n[bold cyan]Embedding Generation Pipeline[/bold cyan]\n")
device_type, device_name = get_device_info()
device = args.device if args.device != "auto" else None
console.print(f"[green]Device:[/green] {device_name}")
console.print(f"[green]Model:[/green] {args.model}")
console.print(f"[green]Batch Size:[/green] {args.batch_size}")
console.print()
# =================================================================
# Step 2: Load Chunks from JSONL
# =================================================================
console.print(f"[cyan]Loading chunks from {args.input_path}...[/cyan]")
chunks = load_chunks_from_jsonl(args.input_path)
console.print(f"[green]Loaded {len(chunks):,} chunks[/green]\n")
# Handle empty input gracefully
if not chunks:
console.print("[yellow]Warning: No chunks to process. Exiting.[/yellow]")
return 0
# =================================================================
# Step 3: Initialize Encoder and Storage
# =================================================================
# Lazy imports for heavy dependencies
from rag_chatbot.embeddings import (
BGEEncoder,
EmbeddingBatch,
EmbeddingStorage,
)
# Create output directory
args.output_dir.mkdir(parents=True, exist_ok=True)
# Initialize encoder with configured model and device
encoder = BGEEncoder(
model_name=args.model,
device=device,
normalize_text=False, # Already normalized during chunk loading
)
# Initialize storage
storage = EmbeddingStorage(args.output_dir)
# =================================================================
# Step 4: Generate Embeddings with Progress Tracking
# =================================================================
# Calculate total batches for progress bar
import math
total_batches = math.ceil(len(chunks) / args.batch_size)
with Progress(
SpinnerColumn(),
TextColumn("[progress.description]{task.description}"),
BarColumn(),
TaskProgressColumn(),
MofNCompleteColumn(),
TimeElapsedColumn(),
TimeRemainingColumn(),
console=console,
) as progress:
# Embedding task
embed_task = progress.add_task(
"[cyan]Embedding chunks...",
total=total_batches,
)
embedding_start = time.perf_counter()
# Generate embeddings
records = create_embedding_records(
chunks=chunks,
encoder=encoder,
batch_size=args.batch_size,
progress=progress,
task_id=embed_task,
)
embedding_time = time.perf_counter() - embedding_start
# Complete the embedding task
progress.update(embed_task, completed=total_batches)
# =================================================================
# Step 5: Save Embeddings to Storage
# =================================================================
save_task = progress.add_task("[cyan]Saving embeddings...", total=1)
# Create EmbeddingBatch
batch = EmbeddingBatch(
model_name=args.model,
dimension=encoder.embedding_dim,
dtype="float16",
records=records,
)
# Save to storage
storage.save(batch)
progress.update(save_task, completed=1)
# =================================================================
# Step 6: Build FAISS and BM25 Indexes
# =================================================================
faiss_time, bm25_time = build_indexes(
output_dir=args.output_dir,
chunks=chunks,
progress=progress,
)
# =================================================================
# Step 7: Publish to HuggingFace (if requested)
# =================================================================
dataset_url: str | None = None
if args.publish:
dataset_url = publish_to_huggingface(
output_dir=args.output_dir,
chunks=chunks,
model_name=args.model,
embedding_dim=encoder.embedding_dim,
progress=progress,
)
# =================================================================
# Step 8: Print Statistics
# =================================================================
total_time = time.perf_counter() - start_time
print_statistics(
total_chunks=len(chunks),
total_time=total_time,
embedding_time=embedding_time,
faiss_time=faiss_time,
bm25_time=bm25_time,
device_name=device_name,
model_name=args.model,
output_dir=args.output_dir,
dataset_url=dataset_url,
)
except FileNotFoundError as exc:
console.print(f"[bold red]Error:[/bold red] {exc}")
return 1
except ValueError as exc:
console.print(f"[bold red]Error:[/bold red] {exc}")
return 1
except KeyboardInterrupt:
console.print("\n[yellow]Interrupted by user[/yellow]")
return 1
except Exception as exc:
console.print(f"[bold red]Unexpected error:[/bold red] {exc}")
logger.exception("Unexpected error during embedding generation")
return 1
else:
console.print("[bold green]Embedding generation complete![/bold green]\n")
return 0
if __name__ == "__main__":
sys.exit(main())
|