code-llm/dvc.lock at main · leonardcser/code-llm · GitHub

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
schema: '2.0'
stages:
  tokenize:
    cmd: mkdir -p tokenizer/build && cmake -S tokenizer -B tokenizer/build
      -DCMAKE_BUILD_TYPE=Release && cmake --build tokenizer/build -j$(nproc) &&
      ./tokenizer/build/tokenize
    deps:
    - path: tokenizer/CMakeLists.txt
      hash: md5
      md5: 769d2e7797f85b125e6dea339062535a
      size: 6405
    - path: tokenizer/src/lib/dataloader.cpp
      hash: md5
      md5: 53ebf2ee5ada158ea331af08bcf08c69
      size: 2070
    - path: tokenizer/src/lib/dataloader.hpp
      hash: md5
      md5: 8c74b03b4a1f4f941751083457b5f6ca
      size: 600
    - path: tokenizer/src/lib/io.cpp
      hash: md5
      md5: 29106b39a48a20faa8d1973ede05ff93
      size: 6004
    - path: tokenizer/src/lib/io.hpp
      hash: md5
      md5: a8177ebb8c302e566c180770cdc09639
      size: 965
    - path: tokenizer/src/lib/text.cpp
      hash: md5
      md5: 419b587d7e870e7665231a8376ea9481
      size: 2872
    - path: tokenizer/src/lib/text.hpp
      hash: md5
      md5: d3d105e19ac1861ea43655dcf6cbfd6e
      size: 134
    - path: tokenizer/src/lib/threading.cpp
      hash: md5
      md5: 5497a70216e75242a6b969862c92fc3b
      size: 1367
    - path: tokenizer/src/lib/threading.hpp
      hash: md5
      md5: 36ad00dece0ca60fdd9f73d4d3d2e982
      size: 1134
    - path: tokenizer/src/lib/tokenizer.cpp
      hash: md5
      md5: beb1595474211bafbde00a5e8946cd79
      size: 43806
    - path: tokenizer/src/lib/tokenizer.hpp
      hash: md5
      md5: 077446b68fa353a79c6af9e80c0926f6
      size: 3706
    - path: tokenizer/src/tokenize.cpp
      hash: md5
      md5: e38e0faff33642ec781675be7ce5cbba
      size: 12728
    params:
      params.yaml:
        tokenize:
          dataset_path: data/bigcode-the-stack-dedup
          glob_pattern: '*.py'
          seed: 42
          max_unique_words: 0
          vocab_size: 35263
          pattern: "'(?i:[sdmt]|ll|ve|re)| ?[A-Za-z_(][A-Za-z_.]*|%(?:\\.\\d+)?[sdifFeEgGxXoc%]|[0-9]{1,3}|
            ?[^ %_A-Za-z0-9]+(?: \")?[\\r\\n]*|%|\\s+$|\\s+(?=\\s)|\\s"
          bos_token: <|startoftext|>
          eos_token: <|endoftext|>
          pad_token: <|pad|>
          cursor_token: <|cursor|>
          edit_start_token: <|edit_start|>
          edit_end_token: <|edit_end|>
          max_train_size: 5GB
          chunk_size: 256M
          dataset_dir: out/tokenize/chunks
          tok_file: out/tokenize/tok.bin
    outs:
    - path: out/tokenize
      hash: md5
      md5: 2e65bce054cb499fc7aa433e08e93711.dir
      size: 10751383763
      nfiles: 12
  train:
    cmd: uv run python src/train.py
    deps:
    - path: out/tokenize
      hash: md5
      md5: 7cf9c81e19c2fed79e239bb6fcc993a7.dir
      size: 10754551901
      nfiles: 15
    - path: src/dataloaders/token_dataloader.py
      hash: md5
      md5: eb3d2bfb850c82f33cb0c8192f767e32
      size: 13755
    - path: src/dataloaders/token_datamodule.py
      hash: md5
      md5: b3b7202bb562d39ddddb6e55fdecd333
      size: 7937
    - path: src/models/qwen3.py
      hash: md5
      md5: 78da8bc156449d3aa63cb52e1a7ee0c3
      size: 14115
    - path: src/train.py
      hash: md5
      md5: 7755a6ba951b23fd5df210d60877d40a
      size: 8951
    - path: src/trainers/trainer.py
      hash: md5
      md5: 2a28aa60aeb22603b161b343db1170e7
      size: 19407
    params:
      params.yaml:
        data.bos_token_id: 35256
        data.dataset_dir: out/tokenize/chunks
        data.eos_token_id: 35257
        data.max_tokens: 0
        data.num_workers: 8
        data.pad_token_id: 35258
        data.seq_length: 256
        data.split_ratio: 0.99
        model:
          hidden_size: 512
          num_hidden_layers: 10
          num_attention_heads: 32
          num_key_value_heads: 2
          intermediate_size: 2024
          max_position_embeddings: 512
          rope_theta: 10000.0
          attention_dropout: 0.1
          rms_norm_eps: 1e-06
          use_sliding_window: false
          sliding_window: 4096
        tokenize.dataset_dir: out/tokenize/chunks
        tokenize.tok_file: out/tokenize/tok.bin
        tokenize.vocab_size: 35263
        training:
          prefix: qwen3
          batch_size: 32
          epochs: 5
          lr: 0.0001
          weight_decay: 0.1
          grad_clip: 1.0
          gradient_accumulation_steps: 4
          use_amp: true
          compile_mode: max-autotune-no-cudagraphs
          devices: 1
          strategy: auto
          seed: 42
          warmup_steps: 5000
          scheduler_t_max_steps:
          log_every_n_steps: 100
          val_every_n_steps: 5000
          save_dir: out/train/checkpoints
          log_dir: out/train/logs
    outs:
    - path: out/train/checkpoints/best.ckpt
      hash: md5
      md5: 9799630d1b9a05d63cceb8fe6f6ea7d0
      size: 1341455351
    - path: out/train/checkpoints/latest.ckpt
      hash: md5
      md5: 9799630d1b9a05d63cceb8fe6f6ea7d0
      size: 1341455351
    - path: out/train/checkpoints/metrics.json
      hash: md5
      md5: 8a80554c91d9fca8acb82f023de02f11
      size: 3
    - path: out/train/logs
      hash: md5
      md5: 4393757035b8cdea1ab037dc777bd0c3.dir
      size: 748623
      nfiles: 3
  export:
    cmd: uv run python src/export.py
    deps:
    - path: out/train/checkpoints/best.ckpt
      hash: md5
      md5: 9799630d1b9a05d63cceb8fe6f6ea7d0
      size: 1341455351
    - path: src/export.py
      hash: md5
      md5: 02c05e45817ebba881397a31f528a379
      size: 4194
    - path: src/models/qwen3.py
      hash: md5
      md5: 78da8bc156449d3aa63cb52e1a7ee0c3
      size: 14115
    params:
      params.yaml:
        tokenize.bos_token: <|startoftext|>
        tokenize.eos_token: <|endoftext|>
        tokenize.pad_token: <|pad|>
        tokenize.tok_file: out/tokenize/tok.bin
        tokenize.vocab_size: 35263
        training.save_dir: out/train/checkpoints
    outs:
    - path: out/export
      hash: md5
      md5: 724d985afee85aff5b65de502e51cdab.dir
      size: 447117099
      nfiles: 5