exp1-whisper-jasmin-child

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.3635
• Wer: 17.3349

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: 48
• eval_batch_size: 32
• seed: 42
• 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: 81
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0258	0.1838	50	1.1952	37.8166
0.9199	0.3676	100	1.0455	35.2233
0.7782	0.5515	150	0.8236	32.5159
0.5719	0.7353	200	0.5861	29.6004
0.4533	0.9191	250	0.4692	23.8233
0.4438	1.1029	300	0.4243	21.0689
0.436	1.2868	350	0.4051	19.8175
0.4189	1.4706	400	0.3933	19.4552
0.4044	1.6544	450	0.3844	19.2170
0.3686	1.8382	500	0.3781	17.7509
0.3758	2.0221	550	0.3734	17.7408
0.4074	2.2059	600	0.3697	17.5697
0.3681	2.3897	650	0.3669	17.4523
0.3604	2.5735	700	0.3650	17.3617
0.3876	2.7574	750	0.3640	17.3483
0.3905	2.9412	800	0.3635	17.3349

Framework versions

• PEFT 0.16.0
• Transformers 4.52.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2