Model save

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "openai/whisper-medium",
+  "_name_or_path": "openai/whisper-large",
   "activation_dropout": 0.0,
   "activation_function": "gelu",
   "architectures": [
@@ -11,17 +11,17 @@
     50257
   ],
   "bos_token_id": 50257,
-  "d_model": 1024,
-  "decoder_attention_heads": 16,
-  "decoder_ffn_dim": 4096,
+  "d_model": 1280,
+  "decoder_attention_heads": 20,
+  "decoder_ffn_dim": 5120,
   "decoder_layerdrop": 0.0,
-  "decoder_layers": 24,
+  "decoder_layers": 32,
   "decoder_start_token_id": 50258,
   "dropout": 0.0,
-  "encoder_attention_heads": 16,
-  "encoder_ffn_dim": 4096,
+  "encoder_attention_heads": 20,
+  "encoder_ffn_dim": 5120,
   "encoder_layerdrop": 0.0,
-  "encoder_layers": 24,
+  "encoder_layers": 32,
   "eos_token_id": 50257,
   "forced_decoder_ids": null,
   "init_std": 0.02,
@@ -30,7 +30,7 @@
   "max_source_positions": 1500,
   "max_target_positions": 448,
   "model_type": "whisper",
-  "num_hidden_layers": 24,
+  "num_hidden_layers": 32,
   "num_mel_bins": 80,
   "pad_token_id": 50257,
   "scale_embedding": false,

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e01ec9eccb75a5c62b8326712df5945b6f8bf00f6fe6486fc6d4972a986c62a0
-size 3055748571
+oid sha256:7b171162e0f6c3c123f162425aaf1faa210d63e93692be3832d6eb1f2da952b1
+size 6173647530

run.sh

@@ -1,15 +1,15 @@
 python run_speech_recognition_seq2seq_streaming.py \
---model_name_or_path="openai/whisper-medium" \
+--model_name_or_path="openai/whisper-large" \
 --dataset_name="mozilla-foundation/common_voice_11_0" \
 --dataset_config_name="vi" \
 --language="vi" \
 --train_split_name="train+validation" \
 --eval_split_name="test" \
 --model_index_name="Whisper Medium Vietnamese" \
---max_steps="1000" \
+--max_steps="1" \
 --output_dir="./" \
---per_device_train_batch_size="2" \
---per_device_eval_batch_size="2" \
+--per_device_train_batch_size="1" \
+--per_device_eval_batch_size="1" \
 --logging_steps="25" \
 --learning_rate="1e-5" \
 --warmup_steps="100" \
@@ -34,4 +34,5 @@ python run_speech_recognition_seq2seq_streaming.py \
 --do_normalize_eval \
 --streaming \
 --use_auth_token \
---push_to_hub
+--push_to_hub \
+--optim="adafactor"

runs/Dec16_22-32-26_ip-172-16-16-209.ec2.internal/1671230049.1602852/events.out.tfevents.1671230049.ip-172-16-16-209.ec2.internal.12258.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a49fb63cfca9722716afb42f531506a9b654e12df67525792484a87c629dcd05
+size 5885

runs/Dec16_22-32-26_ip-172-16-16-209.ec2.internal/events.out.tfevents.1671230049.ip-172-16-16-209.ec2.internal.12258.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e43d4f39f330c9ed2389759df229943264ffdd32b1743d25581fa654b34c5c5a
+size 4281

runs/Dec16_22-40-12_ip-172-16-16-209.ec2.internal/1671230436.8033047/events.out.tfevents.1671230436.ip-172-16-16-209.ec2.internal.18804.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18da789c906ff6e7173fb9c818dfec1bb2be2488e1c35e74d50b6864438866c1
+size 5886

runs/Dec16_22-40-12_ip-172-16-16-209.ec2.internal/events.out.tfevents.1671230436.ip-172-16-16-209.ec2.internal.18804.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:27acde4ac608584e1408edd1527f4caf55b546e133cbc389e1943166dd202287
+size 4282

runs/Dec16_22-43-26_ip-172-16-16-209.ec2.internal/1671230631.900337/events.out.tfevents.1671230631.ip-172-16-16-209.ec2.internal.21903.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:676abc95158f2282c3e598f121c04df5e06037b09b9c62d262993f550f364249
+size 5886

runs/Dec16_22-43-26_ip-172-16-16-209.ec2.internal/events.out.tfevents.1671230631.ip-172-16-16-209.ec2.internal.21903.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8ff1544ca0236947e9b3ef3f8e9b5c4eed6be0ef26c469076e01d4c9c741a30
+size 4627

tokenizer_config.json

@@ -19,7 +19,7 @@
   },
   "errors": "replace",
   "model_max_length": 1024,
-  "name_or_path": "openai/whisper-medium",
+  "name_or_path": "openai/whisper-large",
   "pad_token": null,
   "processor_class": "WhisperProcessor",
   "return_attention_mask": false,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edcc475e37f3aeeeb18abfe94f27779f97b29d710341376a34327f124244bc83
+oid sha256:830f6ecfa0a963c5e4ff14505bfc4032398db3e6c643bafb761761372b95daa0
 size 3567