File size: 4,694 Bytes
a4e32bc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "f9427918",
   "metadata": {},
   "source": [
    "## Load original and transformers tokenizers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "48a7fddf",
   "metadata": {},
   "outputs": [],
   "source": [
    "from huggingface_hub import hf_hub_download\n",
    "\n",
    "original_path = hf_hub_download(repo_id=\"google/codegemma-1.1-2b\", filename=\"tokenizer.model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "aae9d9de",
   "metadata": {},
   "outputs": [],
   "source": [
    "from gemma.tokenizer import Tokenizer\n",
    "\n",
    "original = Tokenizer(original_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "06b063cf",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import GemmaTokenizer, AutoTokenizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "a584d69f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Fails for \"main\"\n",
    "revision = \"refs/pr/4\"\n",
    "\n",
    "t_fast = AutoTokenizer.from_pretrained(\"google/codegemma-1.1-7b-it\", revision=revision)\n",
    "t_slow = GemmaTokenizer.from_pretrained(\"google/codegemma-1.1-7b-it\", revision=revision)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "72a1c087",
   "metadata": {},
   "outputs": [],
   "source": [
    "for s in [\n",
    "    '<start_of_turn>', '<end_of_turn>', '<mask>',\n",
    "    '<|fim_prefix|>', '<|fim_suffix|>', '<|fim_middle|>', '<|file_separator|>'\n",
    "]:\n",
    "    encoded = original.encode(s, bos=False, eos=False)\n",
    "    assert t_fast.encode(s, add_special_tokens=False) == encoded, f\"Failed: {s}\"\n",
    "    assert t_slow.encode(s, add_special_tokens=False) == encoded, f\"Failed: {s}\"\n",
    "    assert t_fast.decode(encoded) == s, f\"Failed: {s}\"\n",
    "    assert t_slow.decode(encoded) == s, f\"Failed: {s}\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8ab89d7b",
   "metadata": {},
   "source": [
    "## Verify on XNLI (validation split)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "0160405a",
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_dataset\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "a743115c",
   "metadata": {},
   "outputs": [],
   "source": [
    "xnli = load_dataset(\"xnli\", \"all_languages\", split=\"validation\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "9a52691b",
   "metadata": {},
   "outputs": [],
   "source": [
    "def verify(lang, text):\n",
    "    encoded_original = original.encode(text, bos=True, eos=False)\n",
    "    encoded_fast = t_fast.encode(text)\n",
    "    encoded_slow = t_slow.encode(text)\n",
    "    assert encoded_fast == encoded_original, f\"Fast encode error: {lang} - {text}\"\n",
    "    assert encoded_slow == encoded_original, f\"Slow encode error: {lang} - {text}\"\n",
    "    decoded = original.decode(encoded_original)\n",
    "    decoded_fast = t_fast.decode(encoded_fast, skip_special_tokens=True)\n",
    "    decoded_slow = t_slow.decode(encoded_slow, skip_special_tokens=True)\n",
    "    assert decoded_fast == decoded, f\"Fast decode error: {lang} - {text}\"\n",
    "    assert decoded_slow == decoded, f\"Slow decode error: {lang} - {text}\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "f3123ffd",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|β–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆβ–ˆ| 2490/2490 [00:30<00:00, 80.45it/s]\n"
     ]
    }
   ],
   "source": [
    "for p in tqdm(xnli[\"premise\"]):\n",
    "    for lang, text in p.items():\n",
    "        verify(lang, text)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}