File size: 4,348 Bytes
b026625
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
679cd9b
b026625
 
 
 
679cd9b
b026625
 
679cd9b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b026625
 
 
 
 
 
 
 
 
 
1165d92
b026625
 
679cd9b
b026625
 
 
679cd9b
b026625
 
 
 
679cd9b
 
b026625
7479c5d
b026625
7479c5d
b026625
 
 
 
 
 
 
679cd9b
 
b026625
 
 
 
 
 
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
#!/usr/bin/env python
# coding: utf-8
# Copyright 2021 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# This script creates a tiny random model
#
# It will be used then as "hf-internal-testing/tiny-electra"

# ***To build from scratch***
#
# 1. clone sentencepiece into a parent dir
# git clone https://github.com/google/sentencepiece
#
# 2. create a new repo at https://huggingface.co/new
# make sure to choose 'hf-internal-testing' as the Owner
#
# 3. clone
# git clone https://huggingface.co/hf-internal-testing/tiny-electra
# cd tiny-electra

# 4. start with some pre-existing script from one of the https://huggingface.co/hf-internal-testing/ tiny model repos, e.g.
# wget https://huggingface.co/hf-internal-testing/tiny-electra/raw/main/make-xlm-roberta.py
# chmod a+x ./make-tiny-electra.py
# mv ./make-tiny-xlm-roberta.py ./make-tiny-electra.py
#
# 5. automatically rename things from the old names to new ones
# perl -pi -e 's|XLMRoberta|Electra|g' make-tiny-electra.py
# perl -pi -e 's|xlm-roberta|electra|g' make-tiny-electra.py
#
# 6. edit and re-run this script while fixing it up
# ./make-tiny-electra.py
#
# 7. add/commit/push
# git add *
# git commit -m "new tiny model"
# git push

# ***To update***
#
# 1. clone the existing repo
# git clone https://huggingface.co/hf-internal-testing/tiny-electra
# cd tiny-electra
#
# 2. edit and re-run this script after doing whatever changes are needed
# ./make-tiny-electra.py
#
# 3. commit/push
# git commit -m "new tiny model"
# git push

import sys
import os

from transformers import ElectraTokenizerFast, ElectraConfig, ElectraForMaskedLM

mname_orig = "google/electra-small-generator"
mname_tiny = "tiny-electra"


### Tokenizer

# Shrink the orig vocab to keep things small (just enough to tokenize any word, so letters+symbols)
# ElectraTokenizerFast is fully defined by a tokenizer.json, which contains the vocab and the ids, so we just need to truncate it wisely
import subprocess
tokenizer_fast = ElectraTokenizerFast.from_pretrained(mname_orig)
vocab_keep_items = 5120
tmp_dir = f"/tmp/{mname_tiny}"
tokenizer_fast.save_pretrained(tmp_dir)
# resize tokenizer.json (vocab.txt will be automatically resized on save_pretrained)
# perl -pi -e 's|(2999).*|$1}}}|' tokenizer.json # 0-indexed, so vocab_keep_items-1!
closing_pat = "}}}"
cmd = (f"perl -pi -e s|({vocab_keep_items-1}).*|$1{closing_pat}| {tmp_dir}/tokenizer.json").split()
result = subprocess.run(cmd, capture_output=True, text=True)
# reload with modified tokenizer
tokenizer_fast_tiny = ElectraTokenizerFast.from_pretrained(tmp_dir)
# it seems that ElectraTokenizer is not needed and ElectraTokenizerFast does the job


### Config

config_tiny = ElectraConfig.from_pretrained(mname_orig)
print(config_tiny)
# remember to update this to the actual config as each model is different and then shrink the numbers
config_tiny.update(dict(
    embedding_size=64,
    hidden_size=64,
    intermediate_size=64,
    max_position_embeddings=512,
    num_attention_heads=2,
    num_hidden_layers=2,
    vocab_size=vocab_keep_items,
))
print("New config", config_tiny)


### Model

model_tiny = ElectraForMaskedLM(config_tiny)

print(f"{mname_tiny}: num of params {model_tiny.num_parameters()}")
model_tiny.resize_token_embeddings(len(tokenizer_fast_tiny))

# Test
inputs = tokenizer_fast_tiny("The capital of France is [MASK].", return_tensors="pt")
outputs = model_tiny(**inputs)
print("Test with normal tokenizer:", len(outputs.logits[0]))

# Save
model_tiny.half() # makes it smaller
model_tiny.save_pretrained(".")
tokenizer_fast_tiny.save_pretrained(".")

#print(model_tiny)

readme = "README.md"
if not os.path.exists(readme):
    with open(readme, "w") as f:
        f.write(f"This is a {mname_tiny} random model to be used for basic testing.\n")

print(f"Generated {mname_tiny}")