up

run_ctc.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+python ./run_flax_speech_recognition_ctc.py \
+        --model_name_or_path="speech-seq2seq/flax-wav2vec2-large-lv60-scan" \
+        --dataset_name="librispeech_asr" \
+        --dataset_cache_dir="/home/sanchitgandhi/cache/huggingface/datasets/" \
+        --dataset_config_name="all" \
+        --train_split_name="train.clean.100+train.clean.360+train.other.500" \
+        --eval_split_name="validation.clean" \
+        --test_split_name="validation.other+test.clean+test.other" \
+        --preprocessing_num_workers="1" \
+        --output_dir="./flax-wav2vec2-ctc-ls-960h-baseline" \
+        --max_steps="50000" \
+        --eval_steps="10000" \
+        --save_steps="10000" \
+        --learning_rate="3e-4" \
+        --logging_steps="25" \
+        --warmup_steps="500" \
+        --do_eval \
+        --do_train \
+        --do_predict \
+        --overwrite_output_dir \
+        --gradient_checkpointing \
+        --freeze_feature_encoder \
+        --push_to_hub \
+        --use_auth_token \
+        --wandb_project="librispeech_960h" \
+        --wandb_name="flax-wav2vec2-ctc-ls-960h-baseline"