wav2vec2 large xlsr 53 punjabi This model is a fine tuned version of Harveenchadha/vakyansh wav2vec2 punjabi pam 10 on the common voice dataset. It achieves the following results on the evaluation set: Loss: 1.2101 Wer: 0.4939 Cer: 0.2238 Evaluation Commands 1. To evaluate on mozilla foundation/common voice 8 0 with split test Inference With LM Training hyperparameters The following hyperparameters were used during training: learning rate: 0.0003 train batch size: 16 eval batch size: 8 seed: 42 gradient accumulation steps: 2 total train batch size: 32 optimizer: Adam with betas=(0.9,0.999) and epsilon=1e 08 lr scheduler type: linear lr scheduler warmup steps: 200 num epochs: 30 mixed precision training: Native AMP Training results Training Loss Epoch Step Validation Loss Wer Cer : : : : : : : : : : : : 11.0563 3.7 100 1.9492 0.7123 0.3872 1.6715 7.41 200 1.3142 0.6433 0.3086 0.9117 11.11 300 1.2733 0.5657 0.2627 0.666 14.81 400 1.2730 0.5598 0.2534 0.4225 18.52 500 1.2548 0.5300 0.2399 0.3209 22.22 600 1.2166 0.5229 0.2372 0.2678 25.93 700 1.1795 0.5041 0.2276 0.2088 29.63 800 1.2101 0.4939 0.2238 Framework versions Transformers 4.17.0.dev0 Pytorch 1.10.2+cu102 Datasets 1.18.2.dev0…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy