From 3dee02cafdc0c0bdf387b0bbab566426c51601df Mon Sep 17 00:00:00 2001
From: ceyda <15624271+cceyda@users.noreply.github.com>
Date: Tue, 6 Apr 2021 23:50:00 +0000
Subject: [PATCH] fix attention flag

---
 README.md                | 10 +++++++---
 preprocessor_config.json |  2 +-
 pytorch_model.bin        |  2 +-
 tokenizer_config.json    |  2 +-
 vocab.json               | 40 +---------------------------------------
 5 files changed, 11 insertions(+), 45 deletions(-)

diff --git a/README.md b/README.md
index c67f30f..601f0de 100644
--- a/README.md
+++ b/README.md
@@ -23,7 +23,7 @@ model-index:
     metrics:
        - name: Test WER
          type: wer
-         value: 24.91
+         value: 22.60
 ---
 
 # Wav2Vec2-Base-760-Turkish
@@ -102,11 +102,13 @@ test_dataset = test_dataset.map(speech_file_to_array_fn)
 
 # Preprocessing the datasets.
 # We need to read the aduio files as arrays
+
+#Attention mask is not used because the base-model was not trained with it. reference: https://github.com/huggingface/transformers/blob/403d530eec105c0e229fc2b754afdf77a4439def/src/transformers/models/wav2vec2/tokenization_wav2vec2.py#L305
 def evaluate(batch):
     inputs = 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
+        logits = model(inputs.input_values.to("cuda")).logits
 
     pred_ids = torch.argmax(logits, dim=-1)
     batch["pred_strings"] = processor.batch_decode(pred_ids,skip_special_tokens=True)
@@ -117,7 +119,9 @@ result = test_dataset.map(evaluate, batched=True, batch_size=8)
 print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
 ```
 
-**Test Result**: 24.91 % (in progress)
+**Test Results**: 
+- WER: 22.602390
+- CER: 6.054137
 
 
 ## Training
diff --git a/preprocessor_config.json b/preprocessor_config.json
index 0886a48..2120dc9 100644
--- a/preprocessor_config.json
+++ b/preprocessor_config.json
@@ -2,7 +2,7 @@
   "do_normalize": true,
   "feature_size": 1,
   "padding_side": "right",
-  "padding_value": 0.0,
+  "padding_value": 0,
   "return_attention_mask": true,
   "sampling_rate": 16000
 }
diff --git a/pytorch_model.bin b/pytorch_model.bin
index 72adfa1..0c98a17 100644
--- a/pytorch_model.bin
+++ b/pytorch_model.bin
@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7ac7eaad990b15315d1772928ea15b9c77d2e259311b5189f9772b04da157294
+oid sha256:adeefd83b89a25212c0d6c74b43b28e367e54cc7fbce63599927f7bc6d2b8ae9
 size 377691502
diff --git a/tokenizer_config.json b/tokenizer_config.json
index 42e4a14..43772fe 100644
--- a/tokenizer_config.json
+++ b/tokenizer_config.json
@@ -1 +1 @@
-{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "|","special_tokens_map_file": "/home/ceyda/workspace/libs/fairseq/hf_finetuned_output/special_tokens_map.json", "tokenizer_file": null}
\ No newline at end of file
+{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "|"}
\ No newline at end of file
diff --git a/vocab.json b/vocab.json
index 2edf7b2..071cf4d 100644
--- a/vocab.json
+++ b/vocab.json
@@ -1,39 +1 @@
-{"|": 4,
- "p": 5,
- "i": 6,
- "r": 7,
- "n": 8,
- "s": 9,
- "ö": 10,
- "z": 11,
- "l": 12,
- "e": 13,
- "h": 14,
- "â": 15,
- "y": 16,
- "a": 17,
- "k": 18,
- "ı": 19,
- "o": 20,
- "m": 21,
- "ü": 22,
- "g": 23,
- "c": 24,
- "b": 25,
- "ş": 26,
- "d": 27,
- "u": 28,
- "t": 29,
- "ç": 30,
- "ğ": 31,
- "v": 32,
- "f": 33,
- "j": 34,
- "x": 35,
- "w": 36,
- "q": 37,
- "î": 38,
- "<s>": 0,
- "<pad>": 1,
- "</s>": 2,
- "<unk>": 3}
+{"<s>": 0, "<pad>": 1, "</s>": 2, "<unk>": 3, "|": 4, "p": 5, "i": 6, "r": 7, "n": 8, "s": 9, "ö": 10, "z": 11, "l": 12, "e": 13, "h": 14, "â": 15, "y": 16, "a": 17, "k": 18, "ı": 19, "o": 20, "m": 21, "ü": 22, "g": 23, "c": 24, "b": 25, "ş": 26, "d": 27, "u": 28, "t": 29, "ç": 30, "ğ": 31, "v": 32, "f": 33, "j": 34, "x": 35, "w": 36, "q": 37, "î": 38}
\ No newline at end of file