reset

config.json

@@ -3,7 +3,7 @@
     "SpeechEncoderDecoderModel"
   ],
   "decoder": {
-    "_name_or_path": "facebook/bart-large",
+    "_name_or_path": "facebook/bart-base",
     "activation_dropout": 0.1,
     "activation_function": "gelu",
     "add_bias_logits": false,
@@ -20,20 +20,20 @@
     "classif_dropout": 0.1,
     "classifier_dropout": 0.0,
     "cross_attention_hidden_size": null,
-    "d_model": 1024,
-    "decoder_attention_heads": 16,
-    "decoder_ffn_dim": 4096,
+    "d_model": 768,
+    "decoder_attention_heads": 12,
+    "decoder_ffn_dim": 3072,
     "decoder_layerdrop": 0.0,
-    "decoder_layers": 12,
+    "decoder_layers": 6,
     "decoder_start_token_id": 2,
     "diversity_penalty": 0.0,
     "do_sample": false,
     "dropout": 0.1,
     "early_stopping": true,
-    "encoder_attention_heads": 16,
-    "encoder_ffn_dim": 4096,
+    "encoder_attention_heads": 12,
+    "encoder_ffn_dim": 3072,
     "encoder_layerdrop": 0.0,
-    "encoder_layers": 12,
+    "encoder_layers": 6,
     "encoder_no_repeat_ngram_size": 0,
     "eos_token_id": 2,
     "exponential_decay_length_penalty": null,
@@ -61,9 +61,10 @@
     "model_type": "bart",
     "no_repeat_ngram_size": 3,
     "normalize_before": false,
+    "normalize_embedding": true,
     "num_beam_groups": 1,
     "num_beams": 4,
-    "num_hidden_layers": 12,
+    "num_hidden_layers": 6,
     "num_return_sequences": 1,
     "output_attentions": false,
     "output_hidden_states": false,
@@ -106,7 +107,7 @@
     "tokenizer_class": null,
     "top_k": 50,
     "top_p": 1.0,
-    "torch_dtype": null,
+    "torch_dtype": "float32",
     "torchscript": false,
     "typical_p": 1.0,
     "use_bfloat16": false,

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57923adad3ced91b7979e1fea25b8ab67480b6461bf758f8f69ca3f76d6bc098
-size 2278337685
+oid sha256:ca8fd76aa7ed61545e51c6e8f239069cefec1e6c185d200b69b3e7c71bdb84df
+size 1649512941

run_speech_recognition_seq2seq.py

@@ -1 +0,0 @@
-/mnt/lscratch/users/gmenon/transformers/examples/pytorch/speech-recognition/run_speech_recognition_seq2seq.py