andyjzhao commited on
Commit
51e26eb
·
verified ·
1 Parent(s): 723d735

Upload hydra_cfg.yaml with huggingface_hub

Browse files
Files changed (1) hide show
  1. hydra_cfg.yaml +343 -0
hydra_cfg.yaml ADDED
@@ -0,0 +1,343 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ RC_augmentation: false
2
+ _dataset_cfg_lookup:
3
+ dlb_cmp_gm12878:
4
+ eval_split: validation
5
+ hf_path: jzshared/dlb_cmp_gm12878
6
+ label_key: label_ut
7
+ mask_key: mask_ut
8
+ num_workers: 0
9
+ path: data/dlb_cmp_gm12878
10
+ pin_memory: true
11
+ reference_id: hg38
12
+ sequence_format: string
13
+ sequence_key: sequence
14
+ shuffle: true
15
+ test_split: test
16
+ train_split: train
17
+ type: cmp_seq
18
+ dlb_cmp_h1hesc:
19
+ eval_split: validation
20
+ hf_path: jzshared/dlb_cmp_h1hesc
21
+ label_key: label_ut
22
+ mask_key: mask_ut
23
+ num_workers: 0
24
+ path: data/dlb_cmp_h1hesc
25
+ pin_memory: true
26
+ reference_id: hg38
27
+ sequence_format: string
28
+ sequence_key: sequence
29
+ shuffle: true
30
+ test_split: test
31
+ train_split: train
32
+ type: cmp_seq
33
+ dlb_cmp_hct116:
34
+ eval_split: validation
35
+ hf_path: jzshared/dlb_cmp_hct116
36
+ label_key: label_ut
37
+ mask_key: mask_ut
38
+ num_workers: 0
39
+ path: data/dlb_cmp_hct116
40
+ pin_memory: true
41
+ reference_id: hg38
42
+ sequence_format: string
43
+ sequence_key: sequence
44
+ shuffle: true
45
+ test_split: test
46
+ train_split: train
47
+ type: cmp_seq
48
+ dlb_cmp_hff:
49
+ eval_split: validation
50
+ hf_path: jzshared/dlb_cmp_hff
51
+ label_key: label_ut
52
+ mask_key: mask_ut
53
+ num_workers: 0
54
+ path: data/dlb_cmp_hff
55
+ pin_memory: true
56
+ reference_id: hg38
57
+ sequence_format: string
58
+ sequence_key: sequence
59
+ shuffle: true
60
+ test_split: test
61
+ train_split: train
62
+ type: cmp_seq
63
+ dlb_cmp_imr90:
64
+ eval_split: validation
65
+ hf_path: jzshared/dlb_cmp_imr90
66
+ label_key: label_ut
67
+ mask_key: mask_ut
68
+ num_workers: 0
69
+ path: data/dlb_cmp_imr90
70
+ pin_memory: true
71
+ reference_id: hg38
72
+ sequence_format: string
73
+ sequence_key: sequence
74
+ shuffle: true
75
+ test_split: test
76
+ train_split: train
77
+ type: cmp_seq
78
+ gencode128k_basic:
79
+ hf_path: jzshared/gencode128k_basic
80
+ path: data/gencode128k_basic
81
+ type: refseq
82
+ gencode128k_debug:
83
+ hf_path: jzshared/gencode128k_debug
84
+ path: data/gencode128k_debug
85
+ type: refseq
86
+ gencode_human_12.8k:
87
+ hf_path: jzshared/gencode_human_12.8k
88
+ path: data/gencode_human_12.8k
89
+ type: refseq
90
+ gencode_human_128k:
91
+ hf_path: jzshared/gencode_human_128k
92
+ path: data/gencode_human_128k
93
+ type: refseq
94
+ hg38_128k:
95
+ hf_path: jzshared/hg38_cds_anchored_128000
96
+ path: data/hg38_cds_anchored_128000
97
+ type: refseq
98
+ hg38_12k:
99
+ hf_path: jzshared/hg38_12800
100
+ path: data/hg38_cds_anchored_len12800_mincds150_1000000samples
101
+ type: refseq
102
+ hg38_cds_4m:
103
+ hf_path: null
104
+ path: data/hg38_cds_dataset_4m_filtered
105
+ type: refseq
106
+ orca32m_cmp_seq:
107
+ eval_split: validation
108
+ hf_path: jzshared/orca32m_cmp
109
+ label_key: label_ut
110
+ mask_key: mask_ut
111
+ num_workers: 0
112
+ path: data/orca32m_cmp_seq
113
+ pin_memory: true
114
+ reference_id: hg38
115
+ sequence_format: string
116
+ sequence_key: sequence
117
+ shuffle: true
118
+ test_split: test
119
+ train_split: train
120
+ type: cmp_seq
121
+ _unimportant_cfg:
122
+ fields:
123
+ - gpus
124
+ - debug
125
+ - wandb
126
+ - env
127
+ - uid
128
+ - local_rank
129
+ - is_distributed
130
+ - master_port
131
+ - device_type
132
+ - cluster
133
+ - world_size
134
+ - train_dataset
135
+ - eval_datasets
136
+ - user_cfg
137
+ - rank
138
+ - device
139
+ - hf_access_token
140
+ - hf_private
141
+ - hf_repo
142
+ - hf_user
143
+ - hf_token
144
+ - save_every
145
+ - eval_steps
146
+ - save_steps
147
+ - upload_to_hf
148
+ - logging
149
+ - log_every
150
+ - use_wandb
151
+ - project_root
152
+ - version
153
+ postfix:
154
+ - _path
155
+ - _file
156
+ - _dir
157
+ - _alias
158
+ - _prefix
159
+ prefix:
160
+ - _
161
+ alias: GeneZip-CMP-7R-LT-32BPT-100B
162
+ alpha_exp: 1.0
163
+ alpha_max: 0.03
164
+ arch: hnet
165
+ batch_size: 32
166
+ bp_per_token: 32
167
+ cluster: mila
168
+ cmd: python src/scripts/train_genezip_v1.py exp=glm/stage1 data=gencode_human_12.8k
169
+ model=hnet/mamba_64m_2dc use_routing_floor=true use_routing_ceiling=true r_hi=0.3
170
+ r_low=0.3 max_routing_tokens=[5000,2000] max_len=12800 batch_size=32 grad_acc_steps=1
171
+ max_train_steps=30600 eval_steps=500 num_valid_samples=3000 upload_to_hf=true wandb.project=DNAFM_v2
172
+ bp_per_token=32 region_info=promoter8_cds16_utr16_exon16_intron4_nig2_dig4 alias=GeneZip-CMP-7R-LT-32BPT-100B
173
+ use_wandb=true hf_repo=jzshared/GeneZip-CMP-7R-LT-32BPT-100B
174
+ config_path: null
175
+ data: gencode_human_12.8k
176
+ data_alias: ${.data}_${max_len}
177
+ dataset: ${_dataset_cfg_lookup[${data}]}
178
+ device: cuda
179
+ device_type: GPU
180
+ dirs:
181
+ data_cache: ${project_root}/data_cache/
182
+ data_storage: ${project_root}/data/
183
+ hydra: ${project_root}/temp/hydra/
184
+ output: ${project_root}/output/${data_alias}/${alias}/
185
+ temp: ${project_root}/temp/working_dir/${uid}/
186
+ wandb_cache: ${oc.env:WANDB_CACHE_DIR,${project_root}/temp/wandb_cache/}
187
+ epochs: 200
188
+ eval_batch_size: ${batch_size}
189
+ eval_steps: 500
190
+ grad_acc_steps: 1
191
+ hf_private: false
192
+ hf_repo: jzshared/GeneZip-CMP-7R-LT-32BPT-100B
193
+ hf_user: jzshared
194
+ is_distributed: true
195
+ local_rank: 0
196
+ log_every: 10
197
+ logging:
198
+ level: info
199
+ log_wandb_metric_to_stdout: true
200
+ lr: 0.001
201
+ master_port: '49381'
202
+ max_data_samples: null
203
+ max_eval_samples: ${num_valid_samples}
204
+ max_grad_norm: 2.0
205
+ max_len: 12800
206
+ max_length: ${max_len}
207
+ max_routing_tokens:
208
+ - 5000
209
+ - 2000
210
+ max_train_steps: 30600
211
+ min_routing_tokens: 8
212
+ mixed_precision: bf16
213
+ mode: Formal
214
+ model:
215
+ arch: hnet
216
+ name: hnet_mamba_64m_2dc
217
+ model_alias: ${oc.select:model.name,UnknownModel}
218
+ model_cfg:
219
+ arch_layout:
220
+ - m2
221
+ - - m2
222
+ - - m15
223
+ - m2
224
+ - m2
225
+ attn_cfg:
226
+ num_heads:
227
+ - 8
228
+ - 8
229
+ - 12
230
+ rotary_emb_dim:
231
+ - 16
232
+ - 16
233
+ - 24
234
+ window_size:
235
+ - 511
236
+ - 511
237
+ - -1
238
+ d_intermediate:
239
+ - 0
240
+ - 0
241
+ - 2048
242
+ d_model:
243
+ - 512
244
+ - 512
245
+ - 768
246
+ max_routing_tokens: ${max_routing_tokens}
247
+ min_routing_tokens: ${min_routing_tokens}
248
+ n_gpt: 1.0
249
+ r_hi: ${r_hi}
250
+ r_low: ${r_low}
251
+ r_warm_up_end: ${r_warm_up_end}
252
+ r_warm_up_start: ${r_warm_up_start}
253
+ ssm_cfg:
254
+ chunk_size: 256
255
+ d_conv: 4
256
+ d_state: 64
257
+ expand: 2
258
+ head_dim: 64
259
+ tie_embeddings: true
260
+ vocab_size: 12
261
+ name: hnet_base
262
+ num_test_samples: 0
263
+ num_train_samples: 0
264
+ num_valid_samples: 3000
265
+ pretrained_ckpt: null
266
+ project_root: ${hydra:runtime.cwd}
267
+ r_hi: 0.3
268
+ r_low: 0.3
269
+ r_warm_up_end: 750
270
+ r_warm_up_start: 200
271
+ rank: 0
272
+ reference_loss: null
273
+ region_info: promoter8_cds16_utr16_exon16_intron4_nig2_dig4
274
+ region_label_map:
275
+ cds: 1
276
+ dig: 6
277
+ exon: 3
278
+ intron: 4
279
+ nig: 5
280
+ promoter: 0
281
+ utr: 2
282
+ save_steps: 3000
283
+ seed: 0
284
+ source: ${dataset.type}
285
+ stop_steps: 0
286
+ strictness_exp: 1.0
287
+ strictness_max: 1.0
288
+ tokenizer: fast
289
+ train_steps: 9999999
290
+ training:
291
+ adam_beta1: 0.9
292
+ adam_beta2: 0.95
293
+ bf16: true
294
+ dataloader_drop_last: true
295
+ dataloader_num_workers: 1
296
+ disable_tqdm: false
297
+ do_train: true
298
+ eval_steps: ${eval_steps}
299
+ eval_strategy: steps
300
+ gradient_accumulation_steps: ${grad_acc_steps}
301
+ gradient_checkpointing: false
302
+ group_by_length: false
303
+ hnet_initializer_range: 0.02
304
+ hnet_lr_multiplier: null
305
+ label_names:
306
+ - input_ids
307
+ learning_rate: ${lr}
308
+ logging_steps: ${log_every}
309
+ lr_scheduler_type: linear
310
+ max_grad_norm: ${max_grad_norm}
311
+ max_train_steps: ${max_train_steps}
312
+ num_train_epochs: ${epochs}
313
+ output_dir: ${dirs.output}
314
+ overrides: {}
315
+ per_device_eval_batch_size: ${eval_batch_size}
316
+ per_device_train_batch_size: ${batch_size}
317
+ remove_unused_columns: false
318
+ report_to: null
319
+ save_steps: ${save_steps}
320
+ save_strategy: steps
321
+ use_lr_multiplier: true
322
+ warmup_steps: 500
323
+ weight_decay: 0.1
324
+ training_alias: ${mode}_glm_s1_${region_info}_bp${bp_per_token}_aw${warmup_steps}_amax${alpha_max}_smax${strictness_max}_lr${lr}_e${epochs}_ms${max_train_steps}_maxlen${max_len}
325
+ uid: p2nb6whc
326
+ upload_to_hf: true
327
+ use_routing_ceiling: true
328
+ use_routing_floor: true
329
+ use_wandb: true
330
+ valid_test_downsample: null
331
+ version: NA
332
+ wandb:
333
+ dir: ${dirs.wandb_cache}
334
+ entity: ${oc.select:env.vars.wandb_entity,${oc.env:WANDB_ENTITY,null}}
335
+ id: p2nb6whc
336
+ mode: online
337
+ name: GeneZip-CMP-7R-LT-32BPT-100B
338
+ project: DNAFM_v2
339
+ step_metric: null
340
+ tags: []
341
+ url: https://wandb.ai/jzshared/DNAFM_v2/runs/p2nb6whc
342
+ warmup_steps: 0
343
+ world_size: 8