seastar105 commited on
Commit
83d1f38
·
verified ·
1 Parent(s): e3f5edf

Upload folder using huggingface_hub

Browse files
tensorboard/version_0/events.out.tfevents.1740009640.3859b00b003f.2855.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cc381d627a237fee168d14a265e3a74060fcf0aa928fb3e39807fb32a27cfd78
3
+ size 9338
tensorboard/version_1/events.out.tfevents.1740009687.3859b00b003f.3072.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e73cd8b2fb9340aaeeb1e9d8acb460203080a28e9598679be6fe7587434da571
3
+ size 9338
tensorboard/version_2/events.out.tfevents.1740009758.3859b00b003f.3274.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d6f6b53c9c100eaf17b607656aaf157ce12c1025b67d95e50527b73bc2672821
3
+ size 146721
tensorboard/version_2/hparams.yaml ADDED
@@ -0,0 +1,132 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ model:
2
+ name: openai/whisper-large-v3-turbo
3
+ gradient_checkpointing: false
4
+ use_lora: true
5
+ train_config:
6
+ lr: 0.001
7
+ scheduler_type: cosine
8
+ warm_steps: 500
9
+ max_steps: ${trainer.max_steps}
10
+ warmup_steps: 200
11
+ model/params/total: 815431680
12
+ model/params/trainable: 6553600
13
+ model/params/non_trainable: 808878080
14
+ data:
15
+ ksponspeech:
16
+ type: wds
17
+ audio_key: flac
18
+ text_key: tn_text
19
+ shuffle: true
20
+ pattern: hf://datasets/seastar105/ksponspeech-webdataset/train/*.tar
21
+ resampled: true
22
+ train_config:
23
+ datasets:
24
+ ksponspeech:
25
+ type: wds
26
+ audio_key: flac
27
+ text_key: text
28
+ shuffle: true
29
+ pattern: hf://datasets/seastar105/ksponspeech-webdataset/train/*.tar
30
+ resampled: true
31
+ address_speech:
32
+ type: wds
33
+ audio_key: wav
34
+ text_key: text
35
+ shuffle: true
36
+ pattern: hf://datasets/seastar105/aihub-71556-webdataset/Training/*.tar
37
+ resampled: true
38
+ kor_broadcast:
39
+ type: wds
40
+ audio_key: wav
41
+ text_key: text
42
+ shuffle: true
43
+ pattern: hf://datasets/seastar105/aihub-464-webdataset/TS*/*.tar
44
+ resampled: true
45
+ kor_lowq_call:
46
+ type: wds
47
+ audio_key: wav
48
+ text_key: text
49
+ shuffle: true
50
+ pattern: hf://datasets/seastar105/aihub-571-webdataset/data/*.tar
51
+ resampled: true
52
+ batch_size: 32
53
+ num_workers: 8
54
+ pin_memory: true
55
+ sampling_weights:
56
+ - 1.0
57
+ - 3.0
58
+ - 10.0
59
+ - 3.0
60
+ epoch_steps: 40000
61
+ val_config:
62
+ datasets:
63
+ ksponspeech:
64
+ type: wds_local
65
+ audio_key: flac
66
+ text_key: text
67
+ shuffle: false
68
+ pattern: ./data/eval/*.tar
69
+ resampled: false
70
+ batch_size: 8
71
+ num_workers: 1
72
+ pin_memory: false
73
+ sampling_weights: null
74
+ tokenizer_name: ${model.name}
75
+ trainer:
76
+ _target_: lightning.pytorch.trainer.Trainer
77
+ default_root_dir: ${paths.output_dir}
78
+ min_epochs: 1
79
+ max_epochs: 10
80
+ accelerator: gpu
81
+ devices: 1
82
+ check_val_every_n_epoch: 1
83
+ deterministic: false
84
+ max_steps: 5000
85
+ precision: bf16-mixed
86
+ accumulate_grad_batches: 8
87
+ gradient_clip_val: 1.0
88
+ num_nodes: 1
89
+ val_check_interval: 8000
90
+ callbacks:
91
+ model_checkpoint:
92
+ _target_: lightning.pytorch.callbacks.ModelCheckpoint
93
+ dirpath: ${paths.output_dir}/checkpoints
94
+ filename: steps_{step:06d}
95
+ monitor: step
96
+ verbose: false
97
+ save_last: true
98
+ save_top_k: 1
99
+ mode: max
100
+ auto_insert_metric_name: false
101
+ save_weights_only: false
102
+ every_n_train_steps: 2000
103
+ train_time_interval: null
104
+ every_n_epochs: null
105
+ save_on_train_epoch_end: null
106
+ model_summary:
107
+ _target_: lightning.pytorch.callbacks.RichModelSummary
108
+ max_depth: -1
109
+ rich_progress_bar:
110
+ _target_: lightning.pytorch.callbacks.RichProgressBar
111
+ learning_rate_monitor:
112
+ _target_: lightning.pytorch.callbacks.LearningRateMonitor
113
+ logging_interval: step
114
+ extras:
115
+ ignore_warnings: false
116
+ enforce_tags: true
117
+ print_config: true
118
+ task_name: whisper-base
119
+ tags:
120
+ - pflow
121
+ ckpt_path: null
122
+ seed: 998244353
123
+ model_config:
124
+ name: openai/whisper-large-v3-turbo
125
+ gradient_checkpointing: false
126
+ use_lora: true
127
+ train_config:
128
+ lr: 0.001
129
+ scheduler_type: cosine
130
+ warm_steps: 500
131
+ max_steps: 5000
132
+ warmup_steps: 200