faster-whisper-medium-sw

README.md

@@ -1,11 +1,9 @@
 ---
 library_name: transformers
 license: apache-2.0
-base_model: openai/whisper-small
+base_model: openai/whisper-medium
 tags:
 - generated_from_trainer
-metrics:
-- wer
 model-index:
 - name: output
   results: []
@@ -16,11 +14,16 @@ should probably proofread and complete it, then remove this comment. -->
 
 # output
 
-This model is a fine-tuned version of [openai/whisper-small](https://huggingface.co/openai/whisper-small) on an unknown dataset.
+This model is a fine-tuned version of [openai/whisper-medium](https://huggingface.co/openai/whisper-medium) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5610
-- Wer: 0.4369
-- Cer: 0.2391
+- eval_loss: 0.4645
+- eval_wer: 0.2780
+- eval_cer: 0.0891
+- eval_runtime: 480.529
+- eval_samples_per_second: 1.199
+- eval_steps_per_second: 0.15
+- epoch: 1.3879
+- step: 1600
 
 ## Model description
 
@@ -40,37 +43,20 @@ More information needed
 
 The following hyperparameters were used during training:
 - learning_rate: 5e-06
-- train_batch_size: 8
+- train_batch_size: 2
 - eval_batch_size: 8
 - seed: 42
-- gradient_accumulation_steps: 4
-- total_train_batch_size: 32
-- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
+- gradient_accumulation_steps: 2
+- total_train_batch_size: 4
+- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
 - lr_scheduler_type: cosine
-- lr_scheduler_warmup_steps: 1000
-- training_steps: 10000
+- lr_scheduler_warmup_steps: 500
+- training_steps: 5000
 - mixed_precision_training: Native AMP
 
-### Training results
-
-| Training Loss | Epoch  | Step | Validation Loss | Wer    | Cer    |
-|:-------------:|:------:|:----:|:---------------:|:------:|:------:|
-| 0.9794        | 0.7882 | 200  | 1.4261          | 0.7751 | 0.2368 |
-| 0.4482        | 1.5754 | 400  | 0.9341          | 0.5545 | 0.1791 |
-| 0.2978        | 2.3626 | 600  | 0.7387          | 0.4951 | 0.1833 |
-| 0.2087        | 3.1498 | 800  | 0.6379          | 0.6575 | 0.3344 |
-| 0.1782        | 3.9379 | 1000 | 0.5686          | 0.3875 | 0.1540 |
-| 0.1275        | 4.7251 | 1200 | 0.5456          | 0.3472 | 0.1295 |
-| 0.0894        | 5.5123 | 1400 | 0.5294          | 0.4248 | 0.2279 |
-| 0.05          | 6.2995 | 1600 | 0.5384          | 0.3768 | 0.1744 |
-| 0.0471        | 7.0867 | 1800 | 0.5511          | 0.5286 | 0.3003 |
-| 0.0297        | 7.8749 | 2000 | 0.5557          | 0.3863 | 0.1668 |
-| 0.0167        | 8.6621 | 2200 | 0.5610          | 0.4369 | 0.2391 |
-
-
 ### Framework versions
 
-- Transformers 4.57.1
+- Transformers 4.48.0
 - Pytorch 2.8.0+cu126
 - Datasets 3.6.0
-- Tokenizers 0.22.1
+- Tokenizers 0.21.4

config.json

@@ -1,4 +1,5 @@
 {
+  "_name_or_path": "openai/whisper-medium",
   "activation_dropout": 0.0,
   "activation_function": "gelu",
   "apply_spec_augment": false,
@@ -9,18 +10,17 @@
   "begin_suppress_tokens": null,
   "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": 12,
+  "decoder_layers": 24,
   "decoder_start_token_id": 50258,
   "dropout": 0.0,
-  "dtype": "float32",
-  "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": [
     [
@@ -49,12 +49,13 @@
   "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,
-  "transformers_version": "4.57.1",
-  "use_cache": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.48.0",
+  "use_cache": false,
   "use_weighted_layer_sum": false,
   "vocab_size": 51865
 }

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": [
@@ -250,6 +234,8 @@
     49870,
     50254,
     50258,
+    50358,
+    50359,
     50360,
     50361,
     50362
@@ -258,5 +244,5 @@
     "transcribe": 50359,
     "translate": 50358
   },
-  "transformers_version": "4.57.1"
+  "transformers_version": "4.48.0"
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0250c7439015f6f7fdf7046ed3085c52239de560e4628f34661618d07370bbdf
-size 966995080
+oid sha256:2accc30dbb7087dedb0e6da93ec0df02ea66fe1f7cff43b9c73fcb5a3307b1b0
+size 3055544304

runs/Oct26_19-08-48_6822e7700798/events.out.tfevents.1761505738.6822e7700798.1087.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:260ed5c83790e40c044187a76d5cd681aec915519049fd7a8c3ad11a6ed0f9a6
+size 16502

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:55350a2ab7e440e3e8a3bfda778c9ef42b5a0fcf176554ec69d124954ab98900
-size 5969
+oid sha256:b42fcdd4228464785684848906472a811a7029daa8bfd596e123baf04d54a07b
+size 5905