ngarneau/copiste-v3-turbo

Whisper Large French - IA Steno

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the IA Steno dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.0719
• Wer: 6.3890

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• 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: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4166	0.0979	50	0.3083	16.7160
0.3502	0.1958	100	0.2590	15.4488
0.3212	0.2938	150	0.2487	18.2639
0.308	0.3917	200	0.2120	13.8478
0.303	0.4896	250	0.1954	14.1892
0.3083	0.5875	300	0.1750	14.1437
0.2818	0.6854	350	0.1563	14.4548
0.2711	0.7834	400	0.1430	11.2224
0.2534	0.8813	450	0.1331	10.3574
0.2661	0.9792	500	0.1236	10.3270
0.1859	1.0764	550	0.1130	9.2647
0.1639	1.1743	600	0.1076	9.2040
0.1635	1.2722	650	0.1015	8.5818
0.1581	1.3701	700	0.0975	8.8398
0.164	1.4681	750	0.0927	7.9748
0.1615	1.5660	800	0.0886	7.8003
0.1638	1.6639	850	0.0827	7.3223
0.1499	1.7618	900	0.0788	6.8897
0.1465	1.8597	950	0.0746	6.5862
0.1388	1.9576	1000	0.0719	6.3890

Framework versions

• Transformers 4.56.2
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.22.1