Text Generation
Transformers
Safetensors
GGUF
Korean
English
llama
3b
korean
from-scratch
orpo
instruction-tuned
preference-aligned
fp8
b200
Eval Results (legacy)
text-generation-inference
Instructions to use pathcosmos/frankenstallm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use pathcosmos/frankenstallm with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="pathcosmos/frankenstallm")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("pathcosmos/frankenstallm") model = AutoModelForCausalLM.from_pretrained("pathcosmos/frankenstallm") - llama-cpp-python
How to use pathcosmos/frankenstallm with llama-cpp-python:
# !pip install llama-cpp-python from llama_cpp import Llama llm = Llama.from_pretrained( repo_id="pathcosmos/frankenstallm", filename="gguf/frankenstallm-3b-Q4_K_M.gguf", )
output = llm( "Once upon a time,", max_tokens=512, echo=True ) print(output)
- Notebooks
- Google Colab
- Kaggle
- Local Apps
- llama.cpp
How to use pathcosmos/frankenstallm with llama.cpp:
Install from brew
brew install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf pathcosmos/frankenstallm:Q4_K_M # Run inference directly in the terminal: llama-cli -hf pathcosmos/frankenstallm:Q4_K_M
Install from WinGet (Windows)
winget install llama.cpp # Start a local OpenAI-compatible server with a web UI: llama-server -hf pathcosmos/frankenstallm:Q4_K_M # Run inference directly in the terminal: llama-cli -hf pathcosmos/frankenstallm:Q4_K_M
Use pre-built binary
# Download pre-built binary from: # https://github.com/ggerganov/llama.cpp/releases # Start a local OpenAI-compatible server with a web UI: ./llama-server -hf pathcosmos/frankenstallm:Q4_K_M # Run inference directly in the terminal: ./llama-cli -hf pathcosmos/frankenstallm:Q4_K_M
Build from source code
git clone https://github.com/ggerganov/llama.cpp.git cd llama.cpp cmake -B build cmake --build build -j --target llama-server llama-cli # Start a local OpenAI-compatible server with a web UI: ./build/bin/llama-server -hf pathcosmos/frankenstallm:Q4_K_M # Run inference directly in the terminal: ./build/bin/llama-cli -hf pathcosmos/frankenstallm:Q4_K_M
Use Docker
docker model run hf.co/pathcosmos/frankenstallm:Q4_K_M
- LM Studio
- Jan
- vLLM
How to use pathcosmos/frankenstallm with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "pathcosmos/frankenstallm" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "pathcosmos/frankenstallm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/pathcosmos/frankenstallm:Q4_K_M
- SGLang
How to use pathcosmos/frankenstallm with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "pathcosmos/frankenstallm" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "pathcosmos/frankenstallm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "pathcosmos/frankenstallm" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "pathcosmos/frankenstallm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Ollama
How to use pathcosmos/frankenstallm with Ollama:
ollama run hf.co/pathcosmos/frankenstallm:Q4_K_M
- Unsloth Studio new
How to use pathcosmos/frankenstallm with Unsloth Studio:
Install Unsloth Studio (macOS, Linux, WSL)
curl -fsSL https://unsloth.ai/install.sh | sh # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for pathcosmos/frankenstallm to start chatting
Install Unsloth Studio (Windows)
irm https://unsloth.ai/install.ps1 | iex # Run unsloth studio unsloth studio -H 0.0.0.0 -p 8888 # Then open http://localhost:8888 in your browser # Search for pathcosmos/frankenstallm to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for pathcosmos/frankenstallm to start chatting
- Docker Model Runner
How to use pathcosmos/frankenstallm with Docker Model Runner:
docker model run hf.co/pathcosmos/frankenstallm:Q4_K_M
- Lemonade
How to use pathcosmos/frankenstallm with Lemonade:
Pull the model
# Download Lemonade from https://lemonade-server.ai/ lemonade pull pathcosmos/frankenstallm:Q4_K_M
Run and chat with the model
lemonade run user.frankenstallm-Q4_K_M
List all available models
lemonade list
File size: 30,402 Bytes
b3d361d | 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 | """
data/tokenize_extra.py β λμ©λ korean_extra/ λ°μ΄ν°μ
λ³λ ¬ ν ν°ν
HuggingFace datasets disk ν¬λ§·(arrow), parquet, jsonl λ± μΈ κ°μ§ ν¬λ§·μ
μλ κ°μ§νμ¬ SentencePiece ν ν¬λμ΄μ λ‘ ν ν°ννκ³ , κ²°κ³Όλ₯Ό uint16 memmap
(.bin) νμΌλ‘ μ μ₯νλ€. 881 GB μ΄μμ λμ©λ λ°μ΄ν°λ μ€νΈλ¦¬λ°Β·μ²ν¬ λ°©μμΌλ‘
μ²λ¦¬νλ€.
μΆλ ₯ ν¬λ§·μ data/dataset.py PackedDataset / TextDataset κ³Ό μμ ν νΈνλλ
numpy uint16 νλ« λ°°μ΄μ΄λ€.
μ¬μ© μμ:
# λ¨μΌ λλ ν 리
python data/tokenize_extra.py \
--input_dir data/korean_extra/fineweb2_edu_ko \
--output data/fineweb2_train.bin \
--num_proc 8
# korean_extra/ μ 체 μλΈλλ ν 리 μΌκ΄ μ²λ¦¬
python data/tokenize_extra.py \
--input_dir data/korean_extra \
--auto_scan \
--output_dir data \
--num_proc 8
# κ³΅κ° κ²μ¦
python -c "
import numpy as np
d = np.memmap('data/fineweb2_train.bin', dtype='uint16', mode='r')
print(f'μ΄ ν ν°: {len(d):,}')
"
"""
from __future__ import annotations
import argparse
import json
import math
import multiprocessing as mp
import os
import struct
import sys
import time
from pathlib import Path
from typing import Generator, Iterable, Iterator
import numpy as np
from tqdm import tqdm
# ---------------------------------------------------------------------------
# SentencePiece μν¬νΈ (μ νμ β μμΌλ©΄ μ€λ₯ λ©μμ§ μΆλ ₯ ν μ’
λ£)
# ---------------------------------------------------------------------------
try:
import sentencepiece as spm
except ImportError:
print(
"ERROR: sentencepiece ν¨ν€μ§κ° μ€μΉλμ΄ μμ§ μμ΅λλ€.\n"
" pip install sentencepiece λ‘ μ€μΉ ν μ¬μ€ννμΈμ.",
file=sys.stderr,
)
sys.exit(1)
# ---------------------------------------------------------------------------
# datasets μν¬νΈ
# ---------------------------------------------------------------------------
try:
import datasets as hf_datasets
except ImportError:
print(
"ERROR: datasets ν¨ν€μ§κ° μ€μΉλμ΄ μμ§ μμ΅λλ€.\n"
" pip install datasets λ‘ μ€μΉ ν μ¬μ€ννμΈμ.",
file=sys.stderr,
)
sys.exit(1)
# ===========================================================================
# μμ
# ===========================================================================
UINT16_MAX = 65535 # uint16 μ€λ²νλ‘ κ²½κ³
MIN_TOKENS = 100 # μ΅μ ν ν° μ (λ―Έλ§μ΄λ©΄ λ²λ¦Ό)
MAX_TOKENS = 32_768 # μ΅λ ν ν° μ (μ΄κ³ΌλΆμ λ²λ¦Ό)
HANGUL_RE_THRESHOLD = 0.10 # νκΈ λΉμ¨ μ΅μ κΈ°μ€ (μ΄ λ―Έλ§μ΄κ³ νκΈ μλ κ²½μ° λ²λ¦Ό)
CHUNK_TOKENS = 500_000 # memmap μ²ν¬ λ¨μ (tokens)
EOS_TOKEN_PLACEHOLDER = 1 # EOS id β SP κΈ°λ³Έκ°, μ€μ idλ λͺ¨λΈμμ μ½μ
# ---------------------------------------------------------------------------
# νκΈ λΉμ¨ νν°
# ---------------------------------------------------------------------------
# ord λ²μ: κ°(AC00) ~ ν£(D7A3), γ±(3131) ~ γ
£(3163)
_HANGUL_START = 0xAC00
_HANGUL_END = 0xD7A3
def _has_enough_korean_or_english(text: str) -> bool:
"""
νκΈ λ¬Έμ λΉμ¨μ΄ HANGUL_RE_THRESHOLD μ΄μμ΄κ±°λ,
ASCII μνλ²³ λΉμ¨μ΄ 0.3 μ΄μμ΄λ©΄ True λ°ν.
λ λ€ μλ κ²½μ° False (μ€κ΅μ΄, μΌλ³Έμ΄λ§ μλ λ±).
"""
if not text:
return False
total = len(text)
hangul_cnt = sum(1 for ch in text if _HANGUL_START <= ord(ch) <= _HANGUL_END)
if hangul_cnt / total >= HANGUL_RE_THRESHOLD:
return True
ascii_alpha = sum(1 for ch in text if ch.isascii() and ch.isalpha())
if ascii_alpha / total >= 0.30:
return True
return False
# ===========================================================================
# ν ν¬λμ΄μ λνΌ (νλ‘μΈμ€ κ° κ³΅μ λΆκ° β κ° μ컀μμ reload)
# ===========================================================================
class SPTokenizer:
"""SentencePiece λͺ¨λΈμ wrappingν κ°λ¨ν ν ν¬λμ΄μ ."""
def __init__(self, model_path: str) -> None:
self._model_path = model_path
self._sp: spm.SentencePieceProcessor | None = None
# νλ‘μΈμ€ fork ν _spκ° NoneμΈ κ²½μ° lazy load
def _ensure_loaded(self) -> None:
if self._sp is None:
sp = spm.SentencePieceProcessor()
sp.Load(self._model_path)
self._sp = sp
@property
def eos_id(self) -> int:
self._ensure_loaded()
return self._sp.eos_id()
@property
def vocab_size(self) -> int:
self._ensure_loaded()
return self._sp.GetPieceSize()
def encode(self, text: str) -> list[int]:
self._ensure_loaded()
return self._sp.EncodeAsIds(text)
# ===========================================================================
# ν¬λ§· κ°μ§ & μ΄ν°λ μ΄ν°
# ===========================================================================
def _detect_format(input_dir: Path) -> str:
"""
λλ ν 리 λ΄μ©μ λ³΄κ³ ν¬λ§·μ μλ κ°μ§νλ€.
λ°νκ°:
"hf_arrow" β HuggingFace datasets disk ν¬λ§· (dataset_info.json μ‘΄μ¬)
"parquet" β .parquet νμΌμ΄ μμ
"jsonl" β .jsonl λλ .json νμΌμ΄ μμ
"unknown" β μ μ μμ
"""
if not input_dir.is_dir():
raise NotADirectoryError(f"μ
λ ₯ κ²½λ‘κ° λλ ν λ¦¬κ° μλλλ€: {input_dir}")
# HF arrow ν¬λ§· νλ³ β dataset_info.json λλ state.jsonμ΄ μμΌλ©΄ HF ν¬λ§·
if (input_dir / "dataset_info.json").exists():
return "hf_arrow"
if (input_dir / "state.json").exists():
return "hf_arrow"
# μλΈ λλ ν 리 μμ dataset_info.jsonμ΄ μλ κ²½μ° (split ν¬ν¨)
for child in input_dir.iterdir():
if child.is_dir() and (child / "dataset_info.json").exists():
return "hf_arrow"
# parquet νμΌ νμΈ
parquets = list(input_dir.rglob("*.parquet"))
if parquets:
return "parquet"
# jsonl / json νμΌ νμΈ
jsonls = list(input_dir.rglob("*.jsonl")) + list(input_dir.rglob("*.json"))
if jsonls:
return "jsonl"
return "unknown"
def _iter_hf_arrow(
input_dir: Path,
text_col: str,
num_proc: int,
) -> Iterator[str]:
"""HuggingFace datasets disk ν¬λ§·μμ ν
μ€νΈλ₯Ό μ€νΈλ¦¬λ°νλ€."""
print(f" [ν¬λ§·] HuggingFace arrow (disk): {input_dir}")
try:
ds = hf_datasets.load_from_disk(str(input_dir))
except Exception as exc:
# DatasetDictμΌ μ μμ β 'train' split μλ
try:
ds_dict = hf_datasets.load_from_disk(str(input_dir))
if isinstance(ds_dict, hf_datasets.DatasetDict):
splits = list(ds_dict.keys())
print(f" DatasetDict κ°μ§. splits={splits}, 'train' split μ¬μ©.")
ds = ds_dict.get("train", ds_dict[splits[0]])
else:
raise exc
except Exception:
raise RuntimeError(
f"HF arrow ν¬λ§· λ‘λ μ€ν¨: {input_dir}\nμμΈ: {exc}"
) from exc
# μ€μ ν
μ€νΈ μ»¬λΌ μ΄λ¦ μΆμ
col = _resolve_text_col(list(ds.column_names), text_col)
print(f" ν
μ€νΈ 컬λΌ: '{col}', μ΄ ν μ: {len(ds):,}")
for row in ds:
yield row[col]
def _iter_parquet(input_dir: Path, text_col: str) -> Iterator[str]:
"""parquet νμΌμμ ν
μ€νΈλ₯Ό μ€νΈλ¦¬λ°νλ€."""
try:
import pyarrow.parquet as pq # type: ignore
except ImportError:
# datasetsλ‘ fallback
print(" [κ²½κ³ ] pyarrow λ―Έμ€μΉ, datasetsλ‘ parquet λ‘λ μλ...")
files = sorted(input_dir.rglob("*.parquet"))
print(f" [ν¬λ§·] parquet ({len(files)} νμΌ): {input_dir}")
ds = hf_datasets.load_dataset(
"parquet",
data_files={"train": [str(f) for f in files]},
split="train",
streaming=True,
)
col = _resolve_text_col(list(ds.column_names), text_col)
print(f" ν
μ€νΈ 컬λΌ: '{col}'")
for row in ds:
yield row[col]
return
files = sorted(input_dir.rglob("*.parquet"))
print(f" [ν¬λ§·] parquet ({len(files)} νμΌ): {input_dir}")
for fpath in files:
pf = pq.ParquetFile(str(fpath))
cols = pf.schema_arrow.names
col = _resolve_text_col(cols, text_col)
for batch in pf.iter_batches(batch_size=1000, columns=[col]):
for val in batch.column(col):
yield val.as_py() or ""
def _iter_jsonl(input_dir: Path, text_col: str) -> Iterator[str]:
"""jsonl / json νμΌμμ ν
μ€νΈλ₯Ό μ€νΈλ¦¬λ°νλ€."""
files = sorted(input_dir.rglob("*.jsonl")) + sorted(input_dir.rglob("*.json"))
# json νμΌ μ€ jsonlμ΄ μλ κ² μ κ±° (νμΌ μμ²΄κ° dictμΈ κ²½μ°)
print(f" [ν¬λ§·] jsonl ({len(files)} νμΌ): {input_dir}")
for fpath in files:
try:
with open(fpath, "r", encoding="utf-8", errors="replace") as fh:
for line in fh:
line = line.strip()
if not line:
continue
try:
obj = json.loads(line)
except json.JSONDecodeError:
continue
if isinstance(obj, str):
yield obj
elif isinstance(obj, dict):
text = (
obj.get(text_col)
or obj.get("text")
or obj.get("content")
or obj.get("document")
or ""
)
yield str(text)
except Exception as exc:
print(f" [κ²½κ³ ] νμΌ μ½κΈ° μ€ν¨: {fpath} β {exc}", file=sys.stderr)
def _resolve_text_col(columns: list[str], preferred: str) -> str:
"""
μ§μ λ 컬λΌμ΄ μμ κ²½μ°, μΌλ°μ μΈ ν
μ€νΈ μ»¬λΌ μ΄λ¦μ μμλλ‘ νμνλ€.
"""
if preferred in columns:
return preferred
for candidate in ("text", "content", "document", "body", "passage"):
if candidate in columns:
print(
f" [INFO] μ»¬λΌ '{preferred}' λ―Έμ‘΄μ¬ β '{candidate}' μ¬μ©. "
f"(μ 체 컬λΌ: {columns[:10]})"
)
return candidate
# λ§μ§λ§ μλ¨: 첫 λ²μ§Έ λ¬Έμμ΄ μ»¬λΌ
print(
f" [κ²½κ³ ] ν
μ€νΈ 컬λΌμ μ°Ύμ§ λͺ»ν¨. 첫 λ²μ§Έ μ»¬λΌ '{columns[0]}' μ¬μ©.",
file=sys.stderr,
)
return columns[0]
def get_text_iterator(
input_dir: Path,
text_col: str,
num_proc: int,
) -> tuple[str, Iterator[str]]:
"""
ν¬λ§·μ μλ κ°μ§νκ³ μλ§μ ν
μ€νΈ μ΄ν°λ μ΄ν°λ₯Ό λ°ννλ€.
Returns:
(fmt, iterator) fmtμ κ°μ§λ ν¬λ§· λ¬Έμμ΄
"""
fmt = _detect_format(input_dir)
if fmt == "hf_arrow":
return fmt, _iter_hf_arrow(input_dir, text_col, num_proc)
elif fmt == "parquet":
return fmt, _iter_parquet(input_dir, text_col)
elif fmt == "jsonl":
return fmt, _iter_jsonl(input_dir, text_col)
else:
raise RuntimeError(
f"μ§μνμ§ μλ ν¬λ§·μ΄κ±°λ μΈμν μ μμ΅λλ€: {input_dir}\n"
f"μ§μ ν¬λ§·: HuggingFace arrow, parquet, jsonl"
)
# ===========================================================================
# λ¨μΌ νλ‘μΈμ€ ν ν°ν μ컀 (multiprocessing.Poolμμ νΈμΆ)
# ===========================================================================
# μ μ ν ν¬λμ΄μ β κ° μ컀 νλ‘μΈμ€μμ ν λ²λ§ μ΄κΈ°ν
_g_sp: SPTokenizer | None = None
_g_model_path: str = ""
def _worker_init(model_path: str) -> None:
"""μ컀 μ΄κΈ°ν ν¨μ: SentencePiece λͺ¨λΈ λ‘λ."""
global _g_sp, _g_model_path
_g_model_path = model_path
_g_sp = SPTokenizer(model_path)
_g_sp._ensure_loaded()
def _worker_tokenize_batch(texts: list[str]) -> list[list[int]]:
"""
ν
μ€νΈ λ°°μΉλ₯Ό ν ν°ννκ³ νμ§ νν°λ₯Ό μ μ©νλ€.
λ°νκ°: μ ν¨ν ν ν° λ¦¬μ€νΈ λͺ©λ‘ (νν° ν΅κ³Όν κ²λ§)
"""
global _g_sp
results: list[list[int]] = []
for text in texts:
if not text or not isinstance(text, str):
continue
# νμ§ νν°: μΈμ΄
if not _has_enough_korean_or_english(text):
continue
try:
ids = _g_sp.encode(text)
except Exception:
continue
# κΈΈμ΄ νν°
if len(ids) < MIN_TOKENS:
continue
if len(ids) > MAX_TOKENS:
ids = ids[:MAX_TOKENS]
results.append(ids)
return results
# ===========================================================================
# memmap μ²ν¬ κΈ°λ° κΈ°λ‘κΈ°
# ===========================================================================
class MemmapWriter:
"""
uint16 numpy memmap νμΌμ ν ν°μ μ²ν¬ λ¨μλ‘ κΈ°λ‘νλ λνΌ.
μ΄κΈ°μ μμ ν¬κΈ°λ‘ μμ±νκ³ , νμν λ resizeνλ€.
μ΅μ’
μ μΌλ‘ μ€μ κΈ°λ‘λ ν¬κΈ°λ‘ truncateνμ¬ μ μ₯νλ€.
"""
def __init__(self, path: Path, initial_size: int = CHUNK_TOKENS) -> None:
self.path = path
path.parent.mkdir(parents=True, exist_ok=True)
self._alloc = max(initial_size, CHUNK_TOKENS)
self._mm = np.memmap(
str(path), dtype="uint16", mode="w+", shape=(self._alloc,)
)
self._pos = 0
def write(self, tokens: Iterable[int]) -> int:
"""tokensλ₯Ό κΈ°λ‘νκ³ κΈ°λ‘λ ν ν° μλ₯Ό λ°ννλ€."""
arr = np.asarray(list(tokens), dtype=np.uint16)
n = len(arr)
if n == 0:
return 0
needed = self._pos + n
if needed > self._alloc:
# λ λ°° λλ νμν ν¬κΈ° μ€ ν° κ°μΌλ‘ νμ₯
new_alloc = max(self._alloc * 2, needed + CHUNK_TOKENS)
self._mm.flush()
del self._mm
self._alloc = new_alloc
self._mm = np.memmap(
str(self.path), dtype="uint16", mode="r+", shape=(self._alloc,)
)
self._mm[self._pos : self._pos + n] = arr
self._pos += n
return n
def finalize(self) -> int:
"""κΈ°λ‘λ μ€μ ν¬κΈ°λ‘ νμΌμ truncateνκ³ λ«λλ€. μ΄ ν ν° μλ₯Ό λ°ννλ€."""
self._mm.flush()
del self._mm
# μ€μ κΈ°λ‘λ ν¬κΈ°λ‘ truncate
final_bytes = self._pos * 2 # uint16 = 2 bytes
with open(str(self.path), "r+b") as fh:
fh.truncate(final_bytes)
return self._pos
# ===========================================================================
# ν΅μ¬ ν ν°ν νμ΄νλΌμΈ
# ===========================================================================
def tokenize_directory(
input_dir: Path,
output_path: Path,
tokenizer_path: str,
text_col: str = "text",
num_proc: int = 8,
batch_size: int = 512,
eos_between_docs: bool = True,
val_split: float = 0.002,
seed: int = 42,
) -> dict:
"""
λ¨μΌ λλ ν 리λ₯Ό ν ν°ννμ¬ .bin νμΌ(λ€)λ‘ μ μ₯νλ€.
Args:
input_dir: μ
λ ₯ λλ ν 리 (ν¬λ§· μλ κ°μ§)
output_path: μΆλ ₯ .bin νμΌ κ²½λ‘ (νλ ¨ μ
)
tokenizer_path: SentencePiece .model νμΌ κ²½λ‘
text_col: ν
μ€νΈ μ»¬λΌ μ΄λ¦ (arrow/parquetμμ μ¬μ©)
num_proc: λ³λ ¬ μ컀 μ
batch_size: μμ»€λΉ λ°°μΉ ν¬κΈ°
eos_between_docs: λ¬Έμ μ¬μ΄μ EOS ν ν° μ½μ
μ¬λΆ
val_split: κ²μ¦ λΆλ¦¬ λΉμ¨ (0 μ΄λ©΄ val νμΌ μμ± μ ν¨)
seed: μ¬νμ± μλ
Returns:
ν΅κ³ dict (total_tokens, train_tokens, val_tokens, skipped, elapsed_s)
"""
t_start = time.time()
# βββ ν ν¬λμ΄μ λ‘λ (λ©μΈ νλ‘μΈμ€: EOS id νμΈ) βββββββββββββββββββββ
sp_main = SPTokenizer(tokenizer_path)
eos_id = sp_main.eos_id
vocab_size = sp_main.vocab_size
print(f" ν ν¬λμ΄μ : {tokenizer_path}")
print(f" vocab_size={vocab_size:,}, eos_id={eos_id}")
if vocab_size > UINT16_MAX:
print(
f" [κ²½κ³ ] vocab_size({vocab_size}) > {UINT16_MAX} "
f"β uint16 μ€λ²νλ‘ κ°λ₯. 65535 μ΄ν idλ§ μμ .",
file=sys.stderr,
)
# βββ ν¬λ§· κ°μ§ & μ΄ν°λ μ΄ν° μμ± βββββββββββββββββββββββββββββββββββββ
fmt, text_iter = get_text_iterator(input_dir, text_col, num_proc)
print(f" ν¬λ§·: {fmt}")
# βββ μΆλ ₯ κ²½λ‘ μ€μ ββββββββββββββββββββββββββββββββββββββββββββββββββββ
train_path = output_path
val_path: Path | None = None
if val_split > 0:
stem = output_path.stem
if "train" in stem:
val_path = output_path.parent / output_path.name.replace("train", "val")
else:
val_path = output_path.with_name(stem + "_val" + output_path.suffix)
print(f" μΆλ ₯(train): {train_path}")
if val_path:
print(f" μΆλ ₯(val): {val_path}")
# βββ memmap κΈ°λ‘κΈ° μ΄κΈ°ν βββββββββββββββββββββββββββββββββββββββββββββ
writer = MemmapWriter(train_path)
val_writer: MemmapWriter | None = MemmapWriter(val_path) if val_path else None
# βββ multiprocessing Pool μμ± ββββββββββββββββββββββββββββββββββββββββ
pool = mp.Pool(
processes=num_proc,
initializer=_worker_init,
initargs=(tokenizer_path,),
)
total_docs = 0
skipped = 0
total_toks = 0
# numpy rng for deterministic val split
rng = np.random.default_rng(seed)
def _submit_batch(batch_texts: list[str]) -> None:
nonlocal total_docs, skipped, total_toks
# λκΈ° map (λ°°μΉ λ¨μ, μμ»€λ³ μλΈλ°°μΉλ‘ λΆν )
sub_size = max(1, len(batch_texts) // num_proc)
sub_batches = [
batch_texts[i : i + sub_size]
for i in range(0, len(batch_texts), sub_size)
]
results_list = pool.map(_worker_tokenize_batch, sub_batches)
for results in results_list:
for ids in results:
total_docs += 1
n = len(ids)
total_toks += n
# EOS ν ν° μ½μ
if eos_between_docs:
ids_out = ids + [eos_id]
else:
ids_out = ids
# val split: 무μμλ‘ val_split λΉμ¨λ§νΌ val νμΌλ‘
if val_writer is not None and rng.random() < val_split:
val_writer.write(ids_out)
else:
writer.write(ids_out)
skipped_in_batch = sum(1 for _ in results) - len(results)
# βββ λ°°μΉ μμ§ & tqdm μ§νλ₯ βββββββββββββββββββββββββββββββββββββββββ
batch_buf: list[str] = []
pbar = tqdm(desc=f"ν ν°ν [{input_dir.name}]", unit="doc", dynamic_ncols=True)
for text in text_iter:
batch_buf.append(text)
if len(batch_buf) >= batch_size * num_proc:
_submit_batch(batch_buf)
pbar.update(len(batch_buf))
pbar.set_postfix(
tokens=f"{total_toks:,}",
docs=f"{total_docs:,}",
refresh=False,
)
batch_buf = []
# λ§μ§λ§ μμ¬ λ°°μΉ μ²λ¦¬
if batch_buf:
_submit_batch(batch_buf)
pbar.update(len(batch_buf))
pbar.close()
pool.close()
pool.join()
# βββ νμΌ λ§λ¬΄λ¦¬ ββββββββββββββββββββββββββββββββββββββββββββββββββββββ
train_tokens = writer.finalize()
val_tokens = val_writer.finalize() if val_writer else 0
elapsed = time.time() - t_start
total_toks_with_eos = train_tokens + val_tokens
print()
print(f" μλ£: {elapsed:.1f}μ΄")
print(f" μ²λ¦¬ λ¬Έμ: {total_docs:,}")
print(f" μ΄ ν ν°(EOS ν¬ν¨): {total_toks_with_eos:,}")
print(f" train: {train_tokens:,} ({train_tokens*2/1e9:.2f} GB)")
if val_tokens:
print(f" val: {val_tokens:,} ({val_tokens*2/1e9:.2f} GB)")
throughput = total_toks_with_eos / elapsed if elapsed > 0 else 0
print(f" μ²λ¦¬μ¨: {throughput/1e6:.2f} M token/s")
return {
"total_docs" : total_docs,
"total_tokens" : total_toks_with_eos,
"train_tokens" : train_tokens,
"val_tokens" : val_tokens,
"elapsed_s" : elapsed,
"train_path" : str(train_path),
"val_path" : str(val_path) if val_path else None,
}
# ===========================================================================
# μλΈλλ ν 리 μλ μ€μΊ λͺ¨λ
# ===========================================================================
def auto_scan_and_tokenize(
root_dir: Path,
output_dir: Path,
tokenizer_path: str,
text_col: str,
num_proc: int,
batch_size: int,
val_split: float,
seed: int,
) -> list[dict]:
"""
root_dir μ μ§μ μμ λλ ν 리λ₯Ό μ€μΊνμ¬ κ°κ° ν ν°ννλ€.
κ° μλΈλλ ν 리μ λν΄:
output_dir/korean_extra_{subdir_name}_train.bin μ μμ±νλ€.
"""
children = sorted(p for p in root_dir.iterdir() if p.is_dir())
if not children:
raise RuntimeError(f"μλΈλλ ν λ¦¬κ° μμ΅λλ€: {root_dir}")
print(f"μλ μ€μΊ: {len(children)}κ° μλΈλλ ν 리 λ°κ²¬")
for ch in children:
print(f" - {ch.name}")
print()
all_stats = []
for child in children:
print("=" * 60)
print(f"μ²λ¦¬ μ€: {child}")
print("=" * 60)
safe_name = child.name.replace("/", "_").replace(" ", "_")
out_name = f"korean_extra_{safe_name}_train.bin"
out_path = output_dir / out_name
try:
stats = tokenize_directory(
input_dir = child,
output_path = out_path,
tokenizer_path = tokenizer_path,
text_col = text_col,
num_proc = num_proc,
batch_size = batch_size,
val_split = val_split,
seed = seed,
)
stats["source"] = child.name
all_stats.append(stats)
except Exception as exc:
print(f" [μ€λ₯] {child.name} μ²λ¦¬ μ€ν¨: {exc}", file=sys.stderr)
all_stats.append({"source": child.name, "error": str(exc)})
print()
return all_stats
# ===========================================================================
# CLI
# ===========================================================================
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(
description=(
"korean_extra/ λμ©λ λ°μ΄ν°μ
μ λ³λ ¬ ν ν°ννμ¬ uint16 memmap(.bin) λ‘ μ μ₯. "
"HuggingFace arrow, parquet, jsonl ν¬λ§· μλ κ°μ§."
),
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
)
# μ
λ ₯
parser.add_argument(
"--input_dir",
required=True,
help="ν ν°νν λλ ν 리 κ²½λ‘. --auto_scan μμλ λ£¨νΈ λλ ν 리.",
)
parser.add_argument(
"--auto_scan",
action="store_true",
help=(
"input_dir μ μ§μ μμ λλ ν 리λ₯Ό λͺ¨λ μμ°¨ μ²λ¦¬. "
"μ΄ κ²½μ° --output_dir μ μ§μ ν΄μΌ ν¨."
),
)
parser.add_argument(
"--text_col",
default="text",
help="ν
μ€νΈ μ»¬λΌ μ΄λ¦ (arrow/parquet/jsonl). μλ μΆμ κ°λ₯.",
)
# μΆλ ₯
out_group = parser.add_mutually_exclusive_group()
out_group.add_argument(
"--output",
default=None,
help="μΆλ ₯ .bin νμΌ κ²½λ‘ (λ¨μΌ λλ ν 리 μ²λ¦¬ μ μ¬μ©).",
)
out_group.add_argument(
"--output_dir",
default=None,
help="μΆλ ₯ .bin νμΌλ€μ μ μ₯ν λλ ν 리 (--auto_scan μ μ¬μ©).",
)
# ν ν¬λμ΄μ
parser.add_argument(
"--tokenizer",
default=(
"/PROJECT/0325120031_A/ghong/taketimes/llm-bang"
"/tokenizer/korean_64k.model"
),
help="SentencePiece .model νμΌ κ²½λ‘.",
)
# μ²λ¦¬ μ΅μ
parser.add_argument(
"--num_proc",
type=int,
default=8,
help="λ³λ ¬ μ컀 μ (multiprocessing.Pool).",
)
parser.add_argument(
"--batch_size",
type=int,
default=512,
help="μμ»€λΉ λ°°μΉ ν¬κΈ° (λ¬Έμ μ).",
)
parser.add_argument(
"--val_split",
type=float,
default=0.002,
help="κ²μ¦ λΆλ¦¬ λΉμ¨ (0.0 μ΄λ©΄ val νμΌ λ―Έμμ±).",
)
parser.add_argument(
"--seed",
type=int,
default=42,
help="μ¬νμ± μλ.",
)
parser.add_argument(
"--no_eos",
action="store_true",
help="λ¬Έμ μ¬μ΄μ EOS ν ν°μ μ½μ
νμ§ μλλ€.",
)
args = parser.parse_args()
# κ²μ¦
if not args.auto_scan and args.output is None:
# μλ μΆλ ₯ κ²½λ‘ μμ±
input_name = Path(args.input_dir).name
args.output = str(
Path(args.input_dir).parent.parent
/ f"korean_extra_{input_name}_train.bin"
)
print(f"[INFO] --output λ―Έμ§μ β μλ μ€μ : {args.output}")
if args.auto_scan and args.output_dir is None:
parser.error("--auto_scan μ¬μ© μ --output_dir μ μ§μ ν΄μΌ ν©λλ€.")
return args
def main() -> None:
args = parse_args()
tokenizer_path = args.tokenizer
if not Path(tokenizer_path).exists():
# fallback: μλκ²½λ‘ μλ
fallback = Path(
"/PROJECT/0325120031_A/ghong/taketimes/llm-bang"
"/tokenizer/korean_64k.model"
)
if fallback.exists():
tokenizer_path = str(fallback)
else:
print(
f"ERROR: ν ν¬λμ΄μ νμΌμ μ°Ύμ μ μμ΅λλ€: {tokenizer_path}",
file=sys.stderr,
)
sys.exit(1)
print("=" * 60)
print(" LLM-Bang tokenize_extra.py")
print("=" * 60)
print(f" μ
λ ₯: {args.input_dir}")
print(f" ν ν¬λμ΄μ : {tokenizer_path}")
print(f" num_proc: {args.num_proc}")
print(f" batch_size: {args.batch_size}")
print(f" val_split: {args.val_split}")
print(f" seed: {args.seed}")
print(f" eos: {not args.no_eos}")
print()
if args.auto_scan:
stats_list = auto_scan_and_tokenize(
root_dir = Path(args.input_dir),
output_dir = Path(args.output_dir),
tokenizer_path = tokenizer_path,
text_col = args.text_col,
num_proc = args.num_proc,
batch_size = args.batch_size,
val_split = args.val_split,
seed = args.seed,
)
print("=" * 60)
print(" μ 체 μμ½")
print("=" * 60)
grand_train = 0
grand_val = 0
for s in stats_list:
if "error" in s:
print(f" {s['source']:40s} ERROR: {s['error']}")
else:
t = s.get("train_tokens", 0)
v = s.get("val_tokens", 0)
grand_train += t
grand_val += v
print(
f" {s['source']:40s} "
f"train={t:>14,} val={v:>12,} "
f"({s['elapsed_s']:.0f}s)"
)
print("-" * 60)
print(
f" {'ν©κ³':40s} "
f"train={grand_train:>14,} val={grand_val:>12,}"
)
print(
f"\n μ΄ ν ν°: {grand_train + grand_val:,} "
f"({(grand_train + grand_val) * 2 / 1e9:.2f} GB)"
)
else:
stats = tokenize_directory(
input_dir = Path(args.input_dir),
output_path = Path(args.output),
tokenizer_path = tokenizer_path,
text_col = args.text_col,
num_proc = args.num_proc,
batch_size = args.batch_size,
eos_between_docs = not args.no_eos,
val_split = args.val_split,
seed = args.seed,
)
print()
print("=" * 60)
print(" κ²°κ³Ό μμ½")
print("=" * 60)
print(f" train .bin : {stats['train_path']}")
if stats.get("val_path"):
print(f" val .bin : {stats['val_path']}")
print(f" train ν ν° : {stats['train_tokens']:,}")
print(f" val ν ν° : {stats['val_tokens']:,}")
print(f" μ²λ¦¬ λ¬Έμ : {stats['total_docs']:,}")
print(f" μμ μκ° : {stats['elapsed_s']:.1f}μ΄")
# κ²μ¦: memmap λ‘λ ν
μ€νΈ
print()
print(" [κ²μ¦] memmap λ‘λ ν
μ€νΈ...")
try:
d = np.memmap(stats["train_path"], dtype="uint16", mode="r")
print(f" memmap shape: {d.shape} dtype: {d.dtype}")
print(f" 첫 10 ν ν°: {d[:10].tolist()}")
except Exception as exc:
print(f" [κ²½κ³ ] memmap λ‘λ μ€ν¨: {exc}", file=sys.stderr)
if __name__ == "__main__":
main()
|