Plim/xls-r-300m-fr

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

Model description

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMONVOICE7_0 - FR dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 7.5e-05
• —trainbatchsize: 16
• —evalbatchsize: 16
• —seed: 42
• —gradientaccumulationsteps: 8
• —totaltrainbatch_size: 128
• —optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• —lrschedulertype: linear
• —lrschedulerwarmup_steps: 2000
• —num_epochs: 2.0
• —mixedprecisiontraining: Native AMP

Training results

It achieves the best result on STEP 6000 on the validation set:

• —Loss: 0.2619
• —Wer: 0.2457

Framework versions

• —Transformers 4.17.0.dev0
• —Pytorch 1.10.2+cu102
• —Datasets 1.18.2.dev0
• —Tokenizers 0.11.0

Evaluation Commands

1. 1.To evaluate on mozilla-foundation/common_voice_7 with split test

python eval.py --model_id Plim/xls-r-300m-fr --dataset mozilla-foundation/common_voice_7_0 --config fr --split test

1. 1.To evaluate on speech-recognition-community-v2/dev_data

python eval.py --model_id Plim/xls-r-300m-fr --dataset speech-recognition-community-v2/dev_data --config fr --split validation --chunk_length_s 5.0 --stride_length_s 1.0