初始化项目，由ModelHub XC社区提供模型

Model: imvladikon/wav2vec2-xls-r-300m-hebrew
Source: Original Platform

run_train.sh

@@ -0,0 +1,39 @@
+export CUDA_VISIBLE_DEVICES="0,1"
+
+python -m torch.distributed.launch --nproc_per_node=2 run_train.py \
+	--dataset_name="imvladikon/hebrew_speech_???" \
+	--use_auth_token="???" \
+    --audio_column_name="audio" \
+    --text_column_name="sentence" \
+	--model_name_or_path="imvladikon/wav2vec2-xls-r-300m-hebrew" \
+    --tokenizer_name_or_path="./wav2vec2-xls-r-300m-hebrew" \
+	--output_dir="./wav2vec2-xls-r-300m-hebrew" \
+	--overwrite_output_dir \
+	--evaluation_strategy="steps" \
+	--length_column_name="input_length" \
+	--gradient_checkpointing \
+	--fp16 \
+	--group_by_length \
+	--num_train_epochs="100" \
+	--per_device_train_batch_size="8" \
+	--per_device_eval_batch_size="8" \
+	--gradient_accumulation_steps="4" \
+	--learning_rate="3e-4" \
+	--warmup_steps="1000" \
+	--save_steps="1000" \
+	--eval_steps="1000" \
+	--preprocessing_num_workers="$(nproc)" \
+	--logging_steps="2000" \
+	--layerdrop="0.0" \
+	--activation_dropout="0.1" \
+	--save_total_limit="3" \
+	--freeze_feature_encoder \
+	--feat_proj_dropout="0.0" \
+	--mask_time_prob="0.75" \
+	--mask_time_length="10" \
+	--mask_feature_prob="0.25" \
+	--mask_feature_length="64" \
+	--do_train --do_eval \
+	--print_samples \
+	--use_augmentations \
+    --push_to_hub