parambharat/whisper-base-ta

Whisper Base Ta - Bharat Ramanathan

This model is a fine-tuned version of openai/whisper-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2269
• Wer: 21.7243

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5559	0.1	1000	0.3963	35.3308
0.3891	0.2	2000	0.3146	29.1511
0.3425	0.3	3000	0.2834	25.5930
0.3108	0.1	4000	0.2669	24.7191
0.2866	0.1	5000	0.2596	25.0936
0.2697	0.2	6000	0.2507	24.5943
0.2421	0.05	6500	0.2411	23.0395
0.2425	0.1	7000	0.2370	23.3804
0.2404	0.15	7500	0.2333	22.7959
0.2381	0.2	8000	0.2311	22.9420
0.2429	0.25	8500	0.2305	22.0166
0.2402	0.3	9000	0.2284	22.1140
0.2377	0.35	9500	0.2271	22.0653
0.2389	0.4	10000	0.2269	21.7243

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2