Thecoder3281f/whisper-small-hi-nscpart123-1000

Whisper Small NSC part 1,2,3 (1000 steps) - Jarrett Er

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

• Loss: 0.2060
• Wer: 9.5998

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3689	0.2008	50	1.0720	62.3310
0.6001	0.4016	100	0.5734	27.9340
0.5761	0.6024	150	0.5917	19.3889
0.5189	0.8032	200	0.5905	20.2271
0.5282	1.0040	250	0.5710	19.7134
0.3914	1.2048	300	0.5133	18.0097
0.369	1.4056	350	0.4591	18.4694
0.3881	1.6064	400	0.4335	16.4684
0.3504	1.8072	450	0.3882	16.5765
0.339	2.0080	500	0.3516	14.6566
0.2437	2.2088	550	0.3359	13.7642
0.2482	2.4096	600	0.3126	13.1963
0.2327	2.6104	650	0.2898	12.7907
0.2041	2.8112	700	0.2685	11.7631
0.1824	3.0120	750	0.2538	10.9789
0.1457	3.2129	800	0.2479	10.5733
0.1617	3.4137	850	0.2304	10.9248
0.1571	3.6145	900	0.2167	9.5457
0.1564	3.8153	950	0.2087	9.3835
0.1503	4.0161	1000	0.2060	9.5998

Framework versions

• PEFT 0.14.0
• Transformers 4.45.2
• Pytorch 2.5.1+cu124
• Datasets 3.2.1.dev0
• Tokenizers 0.20.3