basha3la2/whisper-base-de-v2

Whisper Base DE v2 - Alaa Albasha

This model is a fine-tuned version of basha3la2/whisper-base-de on the ASR German Mixed dataset. It achieves the following results on the evaluation set:

• Loss: 0.2225
• Wer: 14.1615

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: 5e-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3295	0.0667	1000	0.2994	17.9237
0.2862	0.1333	2000	0.2807	17.3261
0.2699	0.2	3000	0.2681	16.6313
0.2515	0.2667	4000	0.2593	16.3450
0.2861	0.3333	5000	0.2519	15.7891
0.2420	0.4	6000	0.2460	15.6604
0.2410	0.4667	7000	0.2411	15.3088
0.2486	0.5333	8000	0.2364	15.0777
0.2304	0.6	9000	0.2326	14.8723
0.2325	0.6667	10000	0.2301	14.6350
0.2225	0.7333	11000	0.2274	14.4608
0.2466	0.8	12000	0.2256	14.3146
0.2200	0.8667	13000	0.2239	14.2793
0.2319	0.9333	14000	0.2229	14.2002
0.2508	1.0	15000	0.2225	14.1615

Framework versions

• Transformers 5.16.1
• Pytorch 2.11.0+cu128
• Datasets 5.0.1
• Tokenizers 0.23.1