pere commited on
Commit
ab6e372
1 Parent(s): 1fb701e
.gitignore ADDED
@@ -0,0 +1 @@
 
 
1
+ checkpoint-*/
config.json CHANGED
@@ -1,5 +1,5 @@
1
  {
2
- "_name_or_path": "openai/whisper-base",
3
  "activation_dropout": 0.0,
4
  "activation_function": "gelu",
5
  "architectures": [
@@ -11,17 +11,17 @@
11
  50257
12
  ],
13
  "bos_token_id": 50257,
14
- "d_model": 512,
15
- "decoder_attention_heads": 8,
16
- "decoder_ffn_dim": 2048,
17
  "decoder_layerdrop": 0.0,
18
- "decoder_layers": 6,
19
  "decoder_start_token_id": 50258,
20
  "dropout": 0.0,
21
- "encoder_attention_heads": 8,
22
- "encoder_ffn_dim": 2048,
23
  "encoder_layerdrop": 0.0,
24
- "encoder_layers": 6,
25
  "eos_token_id": 50257,
26
  "forced_decoder_ids": null,
27
  "init_std": 0.02,
@@ -30,7 +30,7 @@
30
  "max_source_positions": 1500,
31
  "max_target_positions": 448,
32
  "model_type": "whisper",
33
- "num_hidden_layers": 6,
34
  "num_mel_bins": 80,
35
  "pad_token_id": 50257,
36
  "scale_embedding": false,
 
1
  {
2
+ "_name_or_path": "openai/whisper-small",
3
  "activation_dropout": 0.0,
4
  "activation_function": "gelu",
5
  "architectures": [
 
11
  50257
12
  ],
13
  "bos_token_id": 50257,
14
+ "d_model": 768,
15
+ "decoder_attention_heads": 12,
16
+ "decoder_ffn_dim": 3072,
17
  "decoder_layerdrop": 0.0,
18
+ "decoder_layers": 12,
19
  "decoder_start_token_id": 50258,
20
  "dropout": 0.0,
21
+ "encoder_attention_heads": 12,
22
+ "encoder_ffn_dim": 3072,
23
  "encoder_layerdrop": 0.0,
24
+ "encoder_layers": 12,
25
  "eos_token_id": 50257,
26
  "forced_decoder_ids": null,
27
  "init_std": 0.02,
 
30
  "max_source_positions": 1500,
31
  "max_target_positions": 448,
32
  "model_type": "whisper",
33
+ "num_hidden_layers": 12,
34
  "num_mel_bins": 80,
35
  "pad_token_id": 50257,
36
  "scale_embedding": false,
run.sh CHANGED
@@ -7,8 +7,9 @@ python run_speech_recognition_seq2seq_streaming.py \
7
  --model_index_name="Whisper Base Norwegian" \
8
  --max_steps="5000" \
9
  --output_dir="./" \
10
- --per_device_train_batch_size="32" \
11
  --per_device_eval_batch_size="32" \
 
12
  --logging_steps="50" \
13
  --learning_rate="1e-5" \
14
  --warmup_steps="500" \
 
7
  --model_index_name="Whisper Base Norwegian" \
8
  --max_steps="5000" \
9
  --output_dir="./" \
10
+ --per_device_train_batch_size="128" \
11
  --per_device_eval_batch_size="32" \
12
+ --lr_scheduler_type="constant_with_warmup" \
13
  --logging_steps="50" \
14
  --learning_rate="1e-5" \
15
  --warmup_steps="500" \
runs/Dec12_08-40-23_dante/1670830840.7423093/events.out.tfevents.1670830840.dante.3225009.1 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12be5a23fd4a55847c306828d1be0539264b86d9f4593d61be6d348257d9d750
3
+ size 5696
runs/Dec12_08-40-23_dante/events.out.tfevents.1670830840.dante.3225009.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6a07b5fec751b76dd400ecbc88406ceb42568f47d56ca489f8782f18124ac1a
3
+ size 4165
runs/Dec12_08-47-16_dante/1670831255.707328/events.out.tfevents.1670831255.dante.3230904.1 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef5197970336869de45d32e740b8c808fd4cfdd5b56c520df13d54ebb0906c0e
3
+ size 5857
runs/Dec12_08-47-16_dante/events.out.tfevents.1670831255.dante.3230904.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:839afafd6ddf575063388fad3b55544ae380474f898dfcafcfd8a7814fd559c4
3
+ size 4254
tokenizer_config.json CHANGED
@@ -18,8 +18,8 @@
18
  "single_word": false
19
  },
20
  "errors": "replace",
21
- "model_max_length": 448,
22
- "name_or_path": "openai/whisper-base",
23
  "pad_token": null,
24
  "processor_class": "WhisperProcessor",
25
  "return_attention_mask": false,
 
18
  "single_word": false
19
  },
20
  "errors": "replace",
21
+ "model_max_length": 1024,
22
+ "name_or_path": "openai/whisper-small",
23
  "pad_token": null,
24
  "processor_class": "WhisperProcessor",
25
  "return_attention_mask": false,