SPEAK-ASR/whisper-si-exp-2

SPEAK-ASR/speak-whisper-small-si-base-model-hindi

This model is a fine-tuned version of pranay-j/whisper-small-hindi on the Whisper Small - Sinhala ASR Fine-Tuned dataset. It achieves the following results on the evaluation set:

• Loss: 0.2242

Model description

• WandB Project ID: toasty-universe-30
• Use pranay-j/whisper-small-hindi as base model
• Only used 1/3 of OpenSLR dataset
• Used LoRA adapters
• Previous name was SPEAK-ASR/speak-whisper-small-si-base-model-hindi

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: 64
• eval_batch_size: 16
• 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: 500
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.5170	0.6527	500	0.5038
0.2936	1.3055	1000	0.2888
0.2858	1.9582	1500	0.2368
0.2248	2.6110	2000	0.2242

Framework versions

• PEFT 0.18.1
• Transformers 5.0.0rc2
• Pytorch 2.9.1+cu128
• Datasets 4.4.2
• Tokenizers 0.22.2