./outs

README.md

@@ -1,7 +1,7 @@
 ---
 library_name: transformers
-license: mit
-base_model: biodatlab/distill-whisper-th-small
+license: apache-2.0
+base_model: biodatlab/whisper-th-medium-combined
 tags:
 - generated_from_trainer
 model-index:
@@ -14,10 +14,10 @@ should probably proofread and complete it, then remove this comment. -->
 
 # outs
 
-This model is a fine-tuned version of [biodatlab/distill-whisper-th-small](https://huggingface.co/biodatlab/distill-whisper-th-small) on an unknown dataset.
+This model is a fine-tuned version of [biodatlab/whisper-th-medium-combined](https://huggingface.co/biodatlab/whisper-th-medium-combined) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5794
-- Cer: 22.7354
+- Loss: 0.5637
+- Cer: 29.9814
 
 ## Model description
 
@@ -49,9 +49,9 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Cer     |
 |:-------------:|:-----:|:----:|:---------------:|:-------:|
-| 0.3046        | 1.0   | 537  | 0.5714          | 37.0843 |
-| 0.1192        | 2.0   | 1074 | 0.5905          | 29.1700 |
-| 0.0471        | 3.0   | 1611 | 0.5794          | 22.7354 |
+| 0.3375        | 1.0   | 537  | 0.5168          | 24.9036 |
+| 0.2035        | 2.0   | 1074 | 0.5558          | 28.2356 |
+| 0.0809        | 3.0   | 1611 | 0.5637          | 29.9814 |
 
 
 ### Framework versions

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "biodatlab/distill-whisper-th-small",
+  "_name_or_path": "biodatlab/whisper-th-medium-combined",
   "activation_dropout": 0.0,
   "activation_function": "gelu",
   "apply_spec_augment": true,
@@ -13,33 +13,33 @@
   ],
   "bos_token_id": 50257,
   "classifier_proj_size": 256,
-  "d_model": 768,
-  "decoder_attention_heads": 12,
-  "decoder_ffn_dim": 3072,
+  "d_model": 1024,
+  "decoder_attention_heads": 16,
+  "decoder_ffn_dim": 4096,
   "decoder_layerdrop": 0.0,
-  "decoder_layers": 4,
+  "decoder_layers": 24,
   "decoder_start_token_id": 50258,
   "dropout": 0.0,
-  "encoder_attention_heads": 12,
-  "encoder_ffn_dim": 3072,
+  "encoder_attention_heads": 16,
+  "encoder_ffn_dim": 4096,
   "encoder_layerdrop": 0.0,
-  "encoder_layers": 12,
+  "encoder_layers": 24,
   "eos_token_id": 50257,
   "forced_decoder_ids": null,
   "init_std": 0.02,
   "is_encoder_decoder": true,
   "mask_feature_length": 64,
   "mask_feature_min_masks": 0,
-  "mask_feature_prob": 1e-05,
+  "mask_feature_prob": 0.1,
   "mask_time_length": 10,
   "mask_time_min_masks": 2,
-  "mask_time_prob": 1e-05,
+  "mask_time_prob": 0.1,
   "max_length": 448,
   "max_source_positions": 1500,
   "max_target_positions": 448,
   "median_filter_width": 7,
   "model_type": "whisper",
-  "num_hidden_layers": 12,
+  "num_hidden_layers": 24,
   "num_mel_bins": 80,
   "pad_token_id": 50257,
   "scale_embedding": false,

generation_config.json

@@ -1,44 +1,28 @@
 {
   "alignment_heads": [
     [
-      5,
-      3
+      13,
+      15
     ],
     [
-      5,
-      9
-    ],
-    [
-      8,
-      0
-    ],
-    [
-      8,
+      15,
       4
     ],
     [
-      8,
-      7
+      15,
+      15
     ],
     [
-      8,
-      8
+      16,
+      1
     ],
     [
-      9,
+      20,
       0
     ],
     [
-      9,
-      7
-    ],
-    [
-      9,
-      9
-    ],
-    [
-      10,
-      5
+      23,
+      4
     ]
   ],
   "begin_suppress_tokens": [

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cce52bbee5aaeec31aa652ec967274a497b0c3b5b4c3e60dc9d7632c97cafc0f
-size 664564920
+oid sha256:f163cc930ff7ae702cf9db5ea3ecc7ca4d771e708a111ef0a7f6b5fe5eb37b0e
+size 3055544304

runs/Oct08_22-51-00_1f4575752976/events.out.tfevents.1728427864.1f4575752976.965.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6c889b6d7108477c967a9a55b239426a64ff002ed15d86acab598e8752f1a4d
+size 10372

tokenizer_config.json

@@ -12985,6 +12985,5 @@
   "processor_class": "WhisperProcessor",
   "return_attention_mask": false,
   "tokenizer_class": "WhisperTokenizer",
-  "unk_token": "<|endoftext|>",
-  "use_fast": true
+  "unk_token": "<|endoftext|>"
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:88ddf1d9023ed0d0a44a60aab42edf6a7bb32fd900ce380a63a0d73798e2e5f5
+oid sha256:36fd56e536c7d72531cae2b80ea823836ac51b338b4a62138a2a3494649243a4
 size 5304