From aaab3794b6aa79acd984c7fc1215a21792bb0dda Mon Sep 17 00:00:00 2001
From: Patrick von Platen <patrick.v.platen@gmail.com>
Date: Tue, 30 Mar 2021 05:29:48 +0000
Subject: [PATCH] Update README.md

---
 README.md | 80 +++++++++++++++++++++++++++----------------------------
 1 file changed, 40 insertions(+), 40 deletions(-)

diff --git a/README.md b/README.md
index fb9c8f0..6d03f2b 100644
--- a/README.md
+++ b/README.md
@@ -23,7 +23,7 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 12.90
+         value: 12.77
 ---
 
 # Wav2Vec2-Large-XLSR-53-German
@@ -123,30 +123,30 @@ processor = Wav2Vec2Processor.from_pretrained("maxidl/wav2vec2-large-xlsr-german
 model = Wav2Vec2ForCTC.from_pretrained("maxidl/wav2vec2-large-xlsr-german")
 model.to("cuda")
 
-chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\“]'
+chars_to_ignore_regex = '[\\,\\?\\.\\!\\-\\;\\:\\"\\“]'
 resampler = torchaudio.transforms.Resample(48_000, 16_000)
 
 # Preprocessing the datasets.
 # We need to read the aduio files as arrays
 def speech_file_to_array_fn(batch):
-	batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
-	speech_array, sampling_rate = torchaudio.load(batch["path"])
-	batch["speech"] = resampler(speech_array).squeeze().numpy()
-	return batch
+\tbatch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
+\tspeech_array, sampling_rate = torchaudio.load(batch["path"])
+\tbatch["speech"] = resampler(speech_array).squeeze().numpy()
+\treturn batch
 
 test_dataset = test_dataset.map(speech_file_to_array_fn)
 
 # Preprocessing the datasets.
 # We need to read the audio files as arrays
 def evaluate(batch):
-	inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+\tinputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
 
-	with torch.no_grad():
-		logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
+\twith torch.no_grad():
+\t\tlogits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
 
-	pred_ids = torch.argmax(logits, dim=-1)
-	batch["pred_strings"] = processor.batch_decode(pred_ids)
-	return batch
+\tpred_ids = torch.argmax(logits, dim=-1)
+\tbatch["pred_strings"] = processor.batch_decode(pred_ids)
+\treturn batch
 
 result = test_dataset.map(evaluate, batched=True, batch_size=8) # batch_size=8 -> requires ~14.5GB GPU memory
 
@@ -176,7 +176,7 @@ print("Total (chunk_size=1000), WER: {:2f}".format(100 * chunked_wer(result["pre
 # Total (chunk=1000), WER: 12.768981
 ```
 
-**Test Result**: WER: 12.90 %
+**Test Result**: WER: 12.77 %
 
 
 ## Training
@@ -187,32 +187,32 @@ The model was trained for 50k steps, taking around 30 hours on a single A100.
 
 The arguments used for training this model are:
 ```
-python run_finetuning.py \
---model_name_or_path="facebook/wav2vec2-large-xlsr-53" \
---dataset_config_name="de" \
---output_dir=./wav2vec2-large-xlsr-german \
---preprocessing_num_workers="16" \
---overwrite_output_dir \
---num_train_epochs="20" \
---per_device_train_batch_size="64" \
---per_device_eval_batch_size="32" \
---learning_rate="1e-4" \
---warmup_steps="500" \
---evaluation_strategy="steps" \
---save_steps="5000" \
---eval_steps="5000" \
---logging_steps="1000" \
---save_total_limit="3" \
---freeze_feature_extractor \
---activation_dropout="0.055" \
---attention_dropout="0.094" \
---feat_proj_dropout="0.04" \
---layerdrop="0.04" \
---mask_time_prob="0.08" \
---gradient_checkpointing="1" \
---fp16 \
---do_train \
---do_eval \
---dataloader_num_workers="16" \
+python run_finetuning.py \\
+--model_name_or_path="facebook/wav2vec2-large-xlsr-53" \\
+--dataset_config_name="de" \\
+--output_dir=./wav2vec2-large-xlsr-german \\
+--preprocessing_num_workers="16" \\
+--overwrite_output_dir \\
+--num_train_epochs="20" \\
+--per_device_train_batch_size="64" \\
+--per_device_eval_batch_size="32" \\
+--learning_rate="1e-4" \\
+--warmup_steps="500" \\
+--evaluation_strategy="steps" \\
+--save_steps="5000" \\
+--eval_steps="5000" \\
+--logging_steps="1000" \\
+--save_total_limit="3" \\
+--freeze_feature_extractor \\
+--activation_dropout="0.055" \\
+--attention_dropout="0.094" \\
+--feat_proj_dropout="0.04" \\
+--layerdrop="0.04" \\
+--mask_time_prob="0.08" \\
+--gradient_checkpointing="1" \\
+--fp16 \\
+--do_train \\
+--do_eval \\
+--dataloader_num_workers="16" \\
 --group_by_length
 ```