ffbpe 0.1.8

Unicode-aware, streaming BPE training and tiktoken-compatible encoding
Documentation 
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

from __future__ import annotations

import argparse
import json
import statistics
import sys
import time
from collections.abc import Callable, Sequence
from pathlib import Path
from typing import Any, cast

from ffbpe import BoundaryMode
from ffbpe import PreTokenizer

from common import DEFAULT_CHUNK_SIZE
from common import SPECIAL_TOKENS
from common import add_report_args
from common import benchmark_metadata
from common import resolve_report_path
from common import write_report


SCRIPT_NAME = "profile_pretokenizer"


def time_call(fn: Callable[[], Any], repeats: int) -> tuple[Any, dict[str, float]]:
  samples = []
  result = None
  for _ in range(repeats):
    started = time.perf_counter()
    result = fn()
    samples.append(time.perf_counter() - started)
  return result, {
    "min_s": min(samples),
    "median_s": statistics.median(samples),
    "mean_s": statistics.mean(samples),
  }


def summarize_words(words: dict[str, int]) -> dict[str, int]:
  return {
    "unique_words": len(words),
    "occurrences": sum(words.values()),
  }


def profile(args: argparse.Namespace) -> dict[str, Any]:
  pretokenizer = PreTokenizer(SPECIAL_TOKENS, SPECIAL_TOKENS[0])
  file_size = args.input.stat().st_size
  boundary = cast(BoundaryMode, args.boundary)

  boundaries, boundary_timing = time_call(
    lambda: pretokenizer.find_chunk_boundaries(
      args.input,
      chunk_size=args.chunk_size,
      boundary=boundary,
    ),
    args.repeats,
  )

  first_segments = boundaries[:args.segments]
  segment_results = []
  for offset, length in first_segments:
    words, timing = time_call(
      lambda offset=offset, length=length: pretokenizer.get_words_from_segment(args.input, offset, length),
      args.repeats,
    )
    segment_results.append({
      "offset": offset,
      "length": length,
      **timing,
      **summarize_words(words),
    })

  full_words, full_timing = time_call(
    lambda: pretokenizer.get_words_from_file(
      args.input,
      chunk_size=args.chunk_size,
      boundary=boundary,
    ),
    args.repeats,
  )

  return {
    "metadata": benchmark_metadata(
      contract="raw_text_unitoken_pretokenizer_profile",
      script_name=SCRIPT_NAME,
      dataset_name=args.dataset_name,
      config_name=args.config_name,
      experiment_name=args.experiment_name,
      notes=[
        "Profiles FFBPE pretokenizer chunk boundary and word inventory phases.",
      ],
    ),
    "source": {
      "input_kind": "raw_text",
      "input": str(args.input),
      "bytes": file_size,
      "chunk_size": args.chunk_size,
      "boundary": boundary,
    },
    "repeats": args.repeats,
    "boundary_count": len(boundaries),
    "find_chunk_boundaries": boundary_timing,
    "segments": segment_results,
    "get_words_from_file": {
      **full_timing,
      **summarize_words(full_words),
    },
  }


def main(argv: Sequence[str] | None = None) -> int:
  parser = argparse.ArgumentParser(description="Profile FFBPE pretokenizer phases on a raw text file.")
  parser.add_argument("input", type=Path)
  parser.add_argument("--chunk-size", type=int, default=DEFAULT_CHUNK_SIZE)
  parser.add_argument("--boundary", choices=["auto", "eot", "line", "utf8"], default="auto")
  parser.add_argument("--segments", type=int, default=4)
  parser.add_argument("--repeats", type=int, default=1)
  add_report_args(parser)
  args = parser.parse_args(argv)

  if args.chunk_size < 1:
    parser.error("--chunk-size must be at least 1")
  if args.segments < 0:
    parser.error("--segments must be non-negative")
  if args.repeats < 1:
    parser.error("--repeats must be at least 1")

  result = profile(args)
  rendered = json.dumps(result, indent=2)
  if not args.quiet:
    print(rendered)
  write_report(resolve_report_path(args, script_name=SCRIPT_NAME), rendered)
  return 0


if __name__ == "__main__":
  raise SystemExit(main(sys.argv[1:]))