Training and evaluation curves for phoneme training

Logs for evaluation and training

num_gpu: 1; batch_size per GPU: 8; gradient checkpointing true, gradient accumulation 4; no group length; fp16 False

num_gpu: 8; batch_size per GPU: 2; no gradient checkpointing; gradient accumulation 2; no group length; fp16 false

num_gpu: 8; batch_size per GPU: 4; no gradient checkpointing; no gradient accumulation; no group length; fp16 true

num_gpu: 1; batch_size per GPU: 16; gradient checkpointing true, gradient accumulation 2; no group length; fp16 True

batch_size per GPU: 4, no gradient checkpointing

Training curves for https://huggingface.co/patrickvonplaten/wav2vec2-large-xlsr-turkish-demo

Fine-tuning of wav2vec2-base on 100h of Librispeech

Fine-Tuning of a pretrained Wav2Vec2-large-lv60 on 100h of clean librispeech data