Training in progress, step 1000

config.json

@@ -2,15 +2,14 @@
   "_name_or_path": "openai/whisper-medium",
   "activation_dropout": 0.0,
   "activation_function": "gelu",
+  "apply_spec_augment": false,
   "architectures": [
     "WhisperForConditionalGeneration"
   ],
   "attention_dropout": 0.0,
-  "begin_suppress_tokens": [
-    220,
-    50257
-  ],
+  "begin_suppress_tokens": null,
   "bos_token_id": 50257,
+  "classifier_proj_size": 256,
   "d_model": 1024,
   "decoder_attention_heads": 16,
   "decoder_ffn_dim": 4096,
@@ -26,16 +25,24 @@
   "forced_decoder_ids": null,
   "init_std": 0.02,
   "is_encoder_decoder": true,
-  "max_length": 448,
+  "mask_feature_length": 10,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.05,
+  "max_length": null,
   "max_source_positions": 1500,
   "max_target_positions": 448,
+  "median_filter_width": 7,
   "model_type": "whisper",
   "num_hidden_layers": 24,
   "num_mel_bins": 80,
   "pad_token_id": 50257,
   "scale_embedding": false,
   "torch_dtype": "float32",
-  "transformers_version": "4.26.0",
+  "transformers_version": "4.46.0.dev0",
   "use_cache": false,
+  "use_weighted_layer_sum": false,
   "vocab_size": 51865
 }

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:89f033c6a1523e7a92a9f2bf559c6505550a389af90b896f6d72c7908da8cd6f
+size 3055544304

requirements.txt

@@ -1,9 +1,10 @@
 torch>=1.7
 torchaudio
-transformers==4.26.0
-datasets>=2.7.1
+git+https://github.com/huggingface/transformers
+git+https://github.com/huggingface/datasets
 librosa
 jiwer
 evaluate>=0.3.0
 more-itertools
 tensorboard
+accelerate>=0.26.0

run.sh

@@ -9,16 +9,16 @@ WANDB_PROJECT=whisper-medium-eu \
 	--model_index_name="Whisper Medium Basque" \
 	--max_steps="8000" \
 	--output_dir="./" \
-	--per_device_train_batch_size="4" \
+	--per_device_train_batch_size="16" \
 	--per_device_eval_batch_size="8" \
 	--gradient_accumulation_steps="1" \
 	--logging_steps="25" \
-	--learning_rate="1e-5" \
+	--learning_rate="6.25e-6" \
 	--warmup_steps="500" \
 	--evaluation_strategy="steps" \
 	--eval_steps="500" \
 	--save_strategy="steps" \
-	--save_steps="500" \
+	--save_steps="1000" \
 	--generation_max_length="225" \
 	--length_column_name="input_length" \
 	--max_duration_in_seconds="30" \

run_speech_recognition_seq2seq_streaming.py

@@ -274,7 +274,7 @@ def load_maybe_streaming_dataset(dataset_name, dataset_config_name, split="train
     if "+" in split:
         # load multiple splits separated by the `+` symbol with streaming mode
         dataset_splits = [
-            load_dataset(dataset_name, dataset_config_name, split=split_name, streaming=streaming, **kwargs)
+            load_dataset(dataset_name, dataset_config_name, split=split_name, streaming=streaming, trust_remote_code=True, **kwargs)
             for split_name in split.split("+")
         ]
         # interleave multiple splits to form one dataset
@@ -282,7 +282,7 @@ def load_maybe_streaming_dataset(dataset_name, dataset_config_name, split="train
         return interleaved_dataset
     else:
         # load a single split *with* streaming mode
-        dataset = load_dataset(dataset_name, dataset_config_name, split=split, streaming=streaming, **kwargs)
+        dataset = load_dataset(dataset_name, dataset_config_name, split=split, streaming=streaming, trust_remote_code=True, **kwargs)
         return dataset
 
 
@@ -357,7 +357,8 @@ def main():
             data_args.dataset_name,
             data_args.dataset_config_name,
             split=data_args.train_split_name,
-            use_auth_token=True if model_args.use_auth_token else None,
+            # xezpeleta
+            #use_auth_token=True if model_args.use_auth_token else None,
             streaming=data_args.streaming,
         )
 
@@ -366,7 +367,7 @@ def main():
             data_args.dataset_name,
             data_args.dataset_config_name,
             split=data_args.eval_split_name,
-            use_auth_token=True if model_args.use_auth_token else None,
+            #use_auth_token=True if model_args.use_auth_token else None,
             streaming=data_args.streaming,
         )
 

tokenizer_config.json

@@ -12976,44 +12976,14 @@
     "<|nocaptions|>",
     "<|notimestamps|>"
   ],
-  "bos_token": {
-    "__type": "AddedToken",
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
+  "bos_token": "<|endoftext|>",
   "clean_up_tokenization_spaces": true,
-  "eos_token": {
-    "__type": "AddedToken",
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
+  "eos_token": "<|endoftext|>",
   "errors": "replace",
   "model_max_length": 1024,
-  "name_or_path": "openai/whisper-medium",
-  "pad_token": {
-    "__type": "AddedToken",
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
+  "pad_token": "<|endoftext|>",
   "processor_class": "WhisperProcessor",
   "return_attention_mask": false,
-  "special_tokens_map_file": "/home/xezpeleta/.cache/huggingface/hub/models--openai--whisper-medium/snapshots/abdf7c39ab9d0397620ccaea8974cc764cd0953e/special_tokens_map.json",
   "tokenizer_class": "WhisperTokenizer",
-  "unk_token": {
-    "__type": "AddedToken",
-    "content": "<|endoftext|>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  }
+  "unk_token": "<|endoftext|>"
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1673d568bf0b0080cbabe15292938f618760ad94bb79a88f657ec45344a354b5
-size 4024
+oid sha256:b300aa3f6d707e204c955909f5ec2cd2d146917c1914bbddf3758ed2ed2af738
+size 5368