remove useless parameters from training config

helpers/model_init_scripts/init_dummy_model.py

@@ -60,5 +60,8 @@ if __name__ == "__main__":
     model.generation_config.max_length = int(30 * model.audio_encoder.config.frame_rate)
     model.generation_config.do_sample = True  # True
     model.generation_config.guidance_scale = 1  # 3.0
+    
+    model.config.pad_token_id = encodec_vocab_size
+    model.config.decoder_start_token_id = encodec_vocab_size+1
 
     model.save_pretrained(os.path.join(args.save_directory, "tiny-model"))

helpers/model_init_scripts/init_model_300M.py

@@ -60,5 +60,8 @@ if __name__ == "__main__":
     model.generation_config.max_length = int(30 * model.audio_encoder.config.frame_rate)
     model.generation_config.do_sample = True  # True
     model.generation_config.guidance_scale = 1  # 3.0
+    
+    model.config.pad_token_id = encodec_vocab_size
+    model.config.decoder_start_token_id = encodec_vocab_size+1
 
     model.save_pretrained(os.path.join(args.save_directory, "parler-tts-untrained-300M/"))

helpers/training_configs/librispeech_tts_r_300M_dummy.json

@@ -37,9 +37,6 @@
 
     "preprocessing_num_workers": 8,
 
-    "pad_token_id": 1024,
-    "decoder_start_token_id": 1025,
-
     "do_train": true,
     "num_train_epochs": 50,
     "gradient_accumulation_steps": 1,

helpers/training_configs/starting_point_0.01.json

@@ -39,9 +39,6 @@
 
     "preprocessing_num_workers": 8,
 
-    "pad_token_id": 1024,
-    "decoder_start_token_id": 1025,
-
     "do_train": true,
     "num_train_epochs": 40,
     "gradient_accumulation_steps": 1,

training/TRAINING.md

@@ -124,8 +124,6 @@ accelerate launch ./training/run_parler_tts_training.py \
     --add_audio_samples_to_wandb true \
     --id_column_name "id" \
     --preprocessing_num_workers 8 \
-    --pad_token_id 1024 \
-    --decoder_start_token_id 1025 \
     --do_train true \
     --num_train_epochs 50 \
     --gradient_accumulation_steps 1 \

training/run_parler_tts_training.py

@@ -235,11 +235,11 @@ class ModelArguments:
         metadata={"help": "Whether to freeze the text encoder."},
     )
     do_sample: bool = field(
-        default=False,
+        default=True,
         metadata={"help": "Whether to do sampling or greedy decoding."},
     )
     temperature: float = field(
-        default=0.4,
+        default=1.0,
         metadata={"help": "Temperature if sampling."},
     )
     max_length: int = field(
@@ -1018,10 +1018,10 @@ def main():
         {
             "pad_token_id": model_args.pad_token_id
             if model_args.pad_token_id is not None
-            else model.config.pad_token_id,
+            else config.pad_token_id,
             "decoder_start_token_id": model_args.decoder_start_token_id
             if model_args.decoder_start_token_id is not None
-            else model.config.decoder_start_token_id,
+            else config.decoder_start_token_id,
         }
     )