upload

config.json

@@ -1,5 +1,4 @@
 {
-  "_name_or_path": "./wav2vec2_tiny_random/",
   "activation_dropout": 0.1,
   "apply_spec_augment": true,
   "architectures": [
@@ -48,5 +47,5 @@
   "num_hidden_layers": 2,
   "pad_token_id": 0,
   "transformers_version": "4.6.0.dev0",
-  "vocab_size": 32
+  "vocab_size": 12
 }

create_model_files.py

@@ -0,0 +1,48 @@
+#!/usr/bin/env python3
+from transformers import Wav2Vec2Processor, Wav2Vec2FeatureExtractor, Wav2Vec2CTCTokenizer, Wav2Vec2Config, Wav2Vec2Model
+import json
+
+
+conf = Wav2Vec2Config()
+
+conf.conv_dim = [64, 64]
+conf.conv_kernel = [40, 40]
+conf.conv_stride = [30, 30]
+conf.num_feat_extract_layers = 2
+
+conf.hidden_size = 64
+conf.num_conv_pos_embeddings = 64
+conf.num_hidden_layers = 2
+conf.vocab_size = 12
+conf.intermediate_size = 128
+conf.num_conv_pos_embedding_groups = 4
+conf.num_attention_heads = 2
+
+model = Wav2Vec2Model(conf)
+
+vocab = {
+    "a": 0,
+    "b": 1,
+    "c": 2,
+    "d": 3,
+    "e": 4,
+    "f": 5,
+    "g": 6,
+    "<s>": 7,
+    "</s>": 8,
+    "<unk>": 9,
+    "<pad>": 10,
+    "|": 11
+}
+
+with open("vocab.json", "w") as f:
+    f.write(json.dumps(vocab, ensure_ascii=False))
+
+
+tok = Wav2Vec2CTCTokenizer("./vocab.json")
+extract = Wav2Vec2FeatureExtractor()
+
+processor = Wav2Vec2Processor(tokenizer=tok, feature_extractor=extract)
+
+processor.save_pretrained("wav2vec2_tiny_random")
+model.save_pretrained("wav2vec2_tiny_random")

preprocessor_config.json

@@ -0,0 +1,9 @@
+{
+  "do_normalize": true,
+  "feature_extractor_type": "Wav2Vec2FeatureExtractor",
+  "feature_size": 1,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a6a22304e400e96de74369757fbebb00638a9499625b05bb09fd9db783a1c315
+oid sha256:ec0c03140661697880dc1345563e692cc762a4a78c17bdce138c2180abfe7691
 size 1232269

run.sh

@@ -0,0 +1,2 @@
+#!/usr/bin/env bash
+python ../transformers/examples/research_projects/wav2vec2/run_asr.py --output_dir=output_dir --num_train_epochs=1 --per_device_train_batch_size=2 --per_device_eval_batch_size=2 --evaluation_strategy=steps --save_steps=500 --eval_steps=100 --logging_steps=50 --learning_rate=5e-4 --warmup_steps=3000 --model_name_or_path=./wav2vec2_tiny_random --dataset_name=patrickvonplaten/librispeech_asr_dummy --dataset_config_name=clean --train_split_name=validation --validation_split_name=validation --orthography=timit --preprocessing_num_workers=1 --group_by_length --freeze_feature_extractor --verbose_logging

special_tokens_map.json

@@ -0,0 +1 @@
+{"bos_token": "<s>", "eos_token": "</s>", "unk_token": "<unk>", "pad_token": "<pad>"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"unk_token": "<unk>", "bos_token": "<s>", "eos_token": "</s>", "pad_token": "<pad>", "do_lower_case": false, "word_delimiter_token": "|"}

vocab.json

@@ -0,0 +1 @@
+{"a": 0, "b": 1, "c": 2, "d": 3, "e": 4, "f": 5, "g": 6, "<s>": 7, "</s>": 8, "<unk>": 9, "<pad>": 10, "|": 11}