"
+ ]
+ },
+ "execution_count": 16,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "import IPython.display as ipd\n",
+ "import numpy as np\n",
+ "import random\n",
+ "\n",
+ "rand_int = random.randint(0, len(common_voice_train)-1)\n",
+ "\n",
+ "print(\"Target text:\", common_voice_train[rand_int][\"sentence\"])\n",
+ "print(\"Input array shape:\", common_voice_train[rand_int][\"audio\"][\"array\"].shape)\n",
+ "print(\"Sampling rate:\", common_voice_train[rand_int][\"audio\"][\"sampling_rate\"])\n",
+ "ipd.Audio(data=common_voice_train[rand_int][\"audio\"][\"array\"], autoplay=False, rate=16000)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 17,
+ "id": "833f0ebe",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# This does not prepare the input for the Transformer model.\n",
+ "# This will resample the data and convert the sentence into indices\n",
+ "# Batch here is just for one entry (row)\n",
+ "def prepare_dataset(batch):\n",
+ " audio = batch[\"audio\"]\n",
+ " \n",
+ " # batched output is \"un-batched\"\n",
+ " batch[\"input_values\"] = processor(audio[\"array\"], sampling_rate=audio[\"sampling_rate\"]).input_values[0]\n",
+ " batch[\"input_length\"] = len(batch[\"input_values\"])\n",
+ " \n",
+ " with processor.as_target_processor():\n",
+ " batch[\"labels\"] = processor(batch[\"sentence\"]).input_ids\n",
+ " return batch"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 18,
+ "id": "06d2bf9a",
+ "metadata": {
+ "collapsed": true,
+ "jupyter": {
+ "outputs_hidden": true
+ }
+ },
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-184b00bfb8aa2e66.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-8b5882414ea77072.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-3db1be060b3bc81d.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-6c1bbaf643bc5b5f.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-54f525e223e6e4ff.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-e1d9a3bde74cc37c.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-63b9566ab8c13005.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-57b75a9a4b1a9660.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-33f2de77aa38e989.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-5008166243ea7b2b.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-dc8dc0c047dc1c05.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-a84df502424a3cb4.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-6a11a4a3280d39b8.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-93315f983e1a2ae9.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-2534e3d140b7f374.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-decaf49f8e8b5be8/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-3ca6f7f1d923400e.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-b5e44329345c21ce.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-12da16342ab2ddc3.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-59cd84dcce2af6d8.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-a16f6b5197fbc4c9.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-d54449708673af7f.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-a3e265dae408f0ce.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-f7d6406bc2011cb2.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-c1f6017df2a13df3.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-1f54c3872acab8c3.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-abbd70cf97c8fbdb.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-9f8e1f3440a42e0c.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-1502759588e87b89.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-453a18daa2a9e9cc.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-a258ce0e41871520.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-25e62e092344aab4.arrow\n",
+ "Loading cached processed dataset at /workspace/.cache/huggingface/datasets/csv/default-2ae3784a8d52f12b/0.0.0/6b9057d9e23d9d8a2f05b985917a0da84d70c5dae3d22ddd8a3f22fb01c69d9e/cache-e38f0aea7790ffdf.arrow\n"
+ ]
+ }
+ ],
+ "source": [
+ "common_voice_train = common_voice_train.map(prepare_dataset, remove_columns=common_voice_train.column_names, num_proc=16)\n",
+ "common_voice_valid = common_voice_valid.map(prepare_dataset, remove_columns=common_voice_valid.column_names, num_proc=16)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 41,
+ "id": "4b600726",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "# In case the dataset is too long which can lead to OOM. We should filter them out.\n",
+ "# max_input_length_in_sec = 5.0\n",
+ "# common_voice_train = common_voice_train.filter(lambda x: x < max_input_length_in_sec * processor.feature_extractor.sampling_rate, input_columns=[\"input_length\"])"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 19,
+ "id": "e661ff49",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "import torch\n",
+ "\n",
+ "from dataclasses import dataclass, field\n",
+ "from typing import Any, Dict, List, Optional, Union\n",
+ "\n",
+ "@dataclass\n",
+ "class DataCollatorCTCWithPadding:\n",
+ " \"\"\"\n",
+ " Data collator that will dynamically pad the inputs received.\n",
+ " Args:\n",
+ " processor (:class:`~transformers.Wav2Vec2Processor`)\n",
+ " The processor used for proccessing the data.\n",
+ " padding (:obj:`bool`, :obj:`str` or :class:`~transformers.tokenization_utils_base.PaddingStrategy`, `optional`, defaults to :obj:`True`):\n",
+ " Select a strategy to pad the returned sequences (according to the model's padding side and padding index)\n",
+ " among:\n",
+ " * :obj:`True` or :obj:`'longest'`: Pad to the longest sequence in the batch (or no padding if only a single\n",
+ " sequence if provided).\n",
+ " * :obj:`'max_length'`: Pad to a maximum length specified with the argument :obj:`max_length` or to the\n",
+ " maximum acceptable input length for the model if that argument is not provided.\n",
+ " * :obj:`False` or :obj:`'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of\n",
+ " different lengths).\n",
+ " \"\"\"\n",
+ "\n",
+ " processor: Wav2Vec2Processor\n",
+ " padding: Union[bool, str] = True\n",
+ "\n",
+ " def __call__(self, features: List[Dict[str, Union[List[int], torch.Tensor]]]) -> Dict[str, torch.Tensor]:\n",
+ " # split inputs and labels since they have to be of different lenghts and need\n",
+ " # different padding methods\n",
+ " input_features = [{\"input_values\": feature[\"input_values\"]} for feature in features]\n",
+ " label_features = [{\"input_ids\": feature[\"labels\"]} for feature in features]\n",
+ "\n",
+ " batch = self.processor.pad(\n",
+ " input_features,\n",
+ " padding=self.padding,\n",
+ " return_tensors=\"pt\",\n",
+ " )\n",
+ "\n",
+ " with self.processor.as_target_processor():\n",
+ " labels_batch = self.processor.pad(\n",
+ " label_features,\n",
+ " padding=self.padding,\n",
+ " return_tensors=\"pt\",\n",
+ " )\n",
+ "\n",
+ " # replace padding with -100 to ignore loss correctly\n",
+ " labels = labels_batch[\"input_ids\"].masked_fill(labels_batch.attention_mask.ne(1), -100)\n",
+ "\n",
+ " batch[\"labels\"] = labels\n",
+ "\n",
+ " return batch"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 20,
+ "id": "84c914d0",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "data_collator = DataCollatorCTCWithPadding(processor=processor, padding=True)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 21,
+ "id": "ccee9d9b",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "wer_metric = load_metric(\"wer\")\n",
+ "# cer_metric = load_metric(\"cer\")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 22,
+ "id": "f68a0d6e",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "def compute_metrics(pred):\n",
+ " pred_logits = pred.predictions\n",
+ " pred_ids = np.argmax(pred_logits, axis=-1)\n",
+ "\n",
+ " pred.label_ids[pred.label_ids == -100] = tokenizer.pad_token_id\n",
+ "\n",
+ " pred_str = tokenizer.batch_decode(pred_ids)\n",
+ " label_str = tokenizer.batch_decode(pred.label_ids, group_tokens=False)\n",
+ "\n",
+ "# print(\"pred : \", pred_ids[0])\n",
+ "# print(\"label: \", pred.label_ids[0])\n",
+ "# print(\"-----------------\")\n",
+ " \n",
+ " wer = wer_metric.compute(predictions=pred_str, references=label_str)\n",
+ "\n",
+ " return {\"wer\": wer}"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 23,
+ "id": "13f2e1d0",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Some weights of the model checkpoint at facebook/wav2vec2-xls-r-1b were not used when initializing Wav2Vec2ForCTC: ['project_hid.weight', 'quantizer.codevectors', 'project_q.weight', 'project_q.bias', 'quantizer.weight_proj.weight', 'quantizer.weight_proj.bias', 'project_hid.bias']\n",
+ "- This IS expected if you are initializing Wav2Vec2ForCTC from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
+ "- This IS NOT expected if you are initializing Wav2Vec2ForCTC from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
+ "Some weights of Wav2Vec2ForCTC were not initialized from the model checkpoint at facebook/wav2vec2-xls-r-1b and are newly initialized: ['lm_head.weight', 'lm_head.bias']\n",
+ "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
+ ]
+ }
+ ],
+ "source": [
+ "from transformers import Wav2Vec2ForCTC\n",
+ "\n",
+ "model = Wav2Vec2ForCTC.from_pretrained(\n",
+ " \"facebook/wav2vec2-xls-r-1b\",\n",
+ " attention_dropout=0.1,\n",
+ " layerdrop=0.0,\n",
+ " feat_proj_dropout=0.0,\n",
+ " mask_time_prob=0.75, \n",
+ " mask_time_length=10,\n",
+ " mask_feature_prob=0.25,\n",
+ " mask_feature_length=64,\n",
+ " ctc_loss_reduction=\"mean\",\n",
+ " pad_token_id=processor.tokenizer.pad_token_id,\n",
+ " vocab_size=len(processor.tokenizer)\n",
+ ")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 24,
+ "id": "e5ce29d3",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "model.freeze_feature_encoder()"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 25,
+ "id": "69d68721",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "from transformers import TrainingArguments\n",
+ "\n",
+ "training_args = TrainingArguments(\n",
+ " output_dir='.',\n",
+ " group_by_length=True,\n",
+ " per_device_train_batch_size=8,\n",
+ " gradient_accumulation_steps=4,\n",
+ " evaluation_strategy=\"steps\",\n",
+ " gradient_checkpointing=True,\n",
+ " fp16=True,\n",
+ " num_train_epochs=75,\n",
+ " save_steps=400,\n",
+ " eval_steps=400,\n",
+ " logging_steps=100,\n",
+ " learning_rate=1e-5,\n",
+ " warmup_steps=2000,\n",
+ " save_total_limit=3,\n",
+ " load_best_model_at_end=True\n",
+ ")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 26,
+ "id": "19a742e3",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Using amp half precision backend\n"
+ ]
+ }
+ ],
+ "source": [
+ "from transformers import Trainer\n",
+ "\n",
+ "trainer = Trainer(\n",
+ " model=model,\n",
+ " data_collator=data_collator,\n",
+ " args=training_args,\n",
+ " compute_metrics=compute_metrics,\n",
+ " train_dataset=common_voice_train,\n",
+ " eval_dataset=common_voice_valid,\n",
+ " tokenizer=processor.feature_extractor,\n",
+ ")"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 27,
+ "id": "ec1cabad",
+ "metadata": {
+ "collapsed": true,
+ "jupyter": {
+ "outputs_hidden": true
+ }
+ },
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "The following columns in the training set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "/opt/conda/lib/python3.8/site-packages/transformers/optimization.py:306: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use thePyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning\n",
+ " warnings.warn(\n",
+ "***** Running training *****\n",
+ " Num examples = 2353\n",
+ " Num Epochs = 75\n",
+ " Instantaneous batch size per device = 8\n",
+ " Total train batch size (w. parallel, distributed & accumulation) = 32\n",
+ " Gradient Accumulation steps = 4\n",
+ " Total optimization steps = 5475\n"
+ ]
+ },
+ {
+ "data": {
+ "text/html": [
+ "\n",
+ " \n",
+ " \n",
+ "
\n",
+ " [5475/5475 5:17:18, Epoch 74/75]\n",
+ "
\n",
+ " \n",
+ " \n",
+ " \n",
+ " Step | \n",
+ " Training Loss | \n",
+ " Validation Loss | \n",
+ " Wer | \n",
+ "
\n",
+ " \n",
+ " \n",
+ " \n",
+ " 400 | \n",
+ " 3.567100 | \n",
+ " 12.021813 | \n",
+ " 1.000000 | \n",
+ "
\n",
+ " \n",
+ " 800 | \n",
+ " 3.515900 | \n",
+ " 10.633655 | \n",
+ " 1.000000 | \n",
+ "
\n",
+ " \n",
+ " 1200 | \n",
+ " 2.454300 | \n",
+ " 1.825636 | \n",
+ " 0.983858 | \n",
+ "
\n",
+ " \n",
+ " 1600 | \n",
+ " 1.943700 | \n",
+ " 1.123718 | \n",
+ " 0.917272 | \n",
+ "
\n",
+ " \n",
+ " 2000 | \n",
+ " 1.696000 | \n",
+ " 0.824568 | \n",
+ " 0.769976 | \n",
+ "
\n",
+ " \n",
+ " 2400 | \n",
+ " 1.534200 | \n",
+ " 0.643288 | \n",
+ " 0.659403 | \n",
+ "
\n",
+ " \n",
+ " 2800 | \n",
+ " 1.450900 | \n",
+ " 0.550019 | \n",
+ " 0.578692 | \n",
+ "
\n",
+ " \n",
+ " 3200 | \n",
+ " 1.347800 | \n",
+ " 0.507016 | \n",
+ " 0.490718 | \n",
+ "
\n",
+ " \n",
+ " 3600 | \n",
+ " 1.309600 | \n",
+ " 0.469202 | \n",
+ " 0.472559 | \n",
+ "
\n",
+ " \n",
+ " 4000 | \n",
+ " 1.253200 | \n",
+ " 0.444805 | \n",
+ " 0.447942 | \n",
+ "
\n",
+ " \n",
+ " 4400 | \n",
+ " 1.229100 | \n",
+ " 0.437418 | \n",
+ " 0.436642 | \n",
+ "
\n",
+ " \n",
+ " 4800 | \n",
+ " 1.196000 | \n",
+ " 0.431371 | \n",
+ " 0.430993 | \n",
+ "
\n",
+ " \n",
+ " 5200 | \n",
+ " 1.186200 | \n",
+ " 0.423944 | \n",
+ " 0.422115 | \n",
+ "
\n",
+ " \n",
+ "
"
+ ],
+ "text/plain": [
+ ""
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ },
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-400\n",
+ "Configuration saved in ./checkpoint-400/config.json\n",
+ "Model weights saved in ./checkpoint-400/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-400/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-2000] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-800\n",
+ "Configuration saved in ./checkpoint-800/config.json\n",
+ "Model weights saved in ./checkpoint-800/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-800/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-2400] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-1200\n",
+ "Configuration saved in ./checkpoint-1200/config.json\n",
+ "Model weights saved in ./checkpoint-1200/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-1200/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-2800] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-1600\n",
+ "Configuration saved in ./checkpoint-1600/config.json\n",
+ "Model weights saved in ./checkpoint-1600/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-1600/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-400] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-2000\n",
+ "Configuration saved in ./checkpoint-2000/config.json\n",
+ "Model weights saved in ./checkpoint-2000/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-2000/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-800] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-2400\n",
+ "Configuration saved in ./checkpoint-2400/config.json\n",
+ "Model weights saved in ./checkpoint-2400/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-2400/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-1200] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-2800\n",
+ "Configuration saved in ./checkpoint-2800/config.json\n",
+ "Model weights saved in ./checkpoint-2800/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-2800/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-1600] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-3200\n",
+ "Configuration saved in ./checkpoint-3200/config.json\n",
+ "Model weights saved in ./checkpoint-3200/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-3200/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-2000] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-3600\n",
+ "Configuration saved in ./checkpoint-3600/config.json\n",
+ "Model weights saved in ./checkpoint-3600/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-3600/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-2400] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-4000\n",
+ "Configuration saved in ./checkpoint-4000/config.json\n",
+ "Model weights saved in ./checkpoint-4000/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-4000/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-2800] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-4400\n",
+ "Configuration saved in ./checkpoint-4400/config.json\n",
+ "Model weights saved in ./checkpoint-4400/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-4400/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-3200] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-4800\n",
+ "Configuration saved in ./checkpoint-4800/config.json\n",
+ "Model weights saved in ./checkpoint-4800/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-4800/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-3600] due to args.save_total_limit\n",
+ "The following columns in the evaluation set don't have a corresponding argument in `Wav2Vec2ForCTC.forward` and have been ignored: input_length.\n",
+ "***** Running Evaluation *****\n",
+ " Num examples = 262\n",
+ " Batch size = 8\n",
+ "Saving model checkpoint to ./checkpoint-5200\n",
+ "Configuration saved in ./checkpoint-5200/config.json\n",
+ "Model weights saved in ./checkpoint-5200/pytorch_model.bin\n",
+ "Configuration saved in ./checkpoint-5200/preprocessor_config.json\n",
+ "Deleting older checkpoint [checkpoint-4000] due to args.save_total_limit\n",
+ "\n",
+ "\n",
+ "Training completed. Do not forget to share your model on huggingface.co/models =)\n",
+ "\n",
+ "\n",
+ "Loading best model from ./checkpoint-5200 (score: 0.4239440858364105).\n"
+ ]
+ },
+ {
+ "data": {
+ "text/plain": [
+ "TrainOutput(global_step=5475, training_loss=2.0858804112264555, metrics={'train_runtime': 19044.9812, 'train_samples_per_second': 9.266, 'train_steps_per_second': 0.287, 'total_flos': 8.13888461360448e+19, 'train_loss': 2.0858804112264555, 'epoch': 74.99})"
+ ]
+ },
+ "execution_count": 27,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "trainer.train()"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 37,
+ "id": "3666e499",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Saving model checkpoint to .\n",
+ "Configuration saved in ./config.json\n",
+ "Model weights saved in ./pytorch_model.bin\n",
+ "Configuration saved in ./preprocessor_config.json\n"
+ ]
+ }
+ ],
+ "source": [
+ "# trainer.save_model('vitouphy/wav2vec2-xls-r-1b-km')\n",
+ "trainer.save_model('.')"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 40,
+ "id": "e52460e6",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "tokenizer config file saved in ./tokenizer_config.json\n",
+ "Special tokens file saved in ./special_tokens_map.json\n",
+ "added tokens file saved in ./added_tokens.json\n"
+ ]
+ },
+ {
+ "data": {
+ "text/plain": [
+ "('./tokenizer_config.json',\n",
+ " './special_tokens_map.json',\n",
+ " './vocab.json',\n",
+ " './added_tokens.json')"
+ ]
+ },
+ "execution_count": 40,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "tokenizer.save_pretrained(\".\")\n",
+ "# tokenizer.push_to_hub('vitouphy/wav2vec2-xls-r-1b-km')"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 41,
+ "id": "49e54770",
+ "metadata": {},
+ "outputs": [],
+ "source": [
+ "kwargs = {\n",
+ " \"finetuned_from\": \"facebook/wav2vec2-xls-r-1b\",\n",
+ " \"tasks\": \"speech-recognition\",\n",
+ " \"tags\": [\"automatic-speech-recognition\", \"openslr\", \"robust-speech-event\", \"km\"],\n",
+ " \"dataset_args\": f\"Config: km, Training split: train, Eval split: validation\",\n",
+ " \"dataset\": \"openslr\",\n",
+ " \"language\": \"km\"\n",
+ "}"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 42,
+ "id": "6ab99e82",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Dropping the following result as it does not have all the necessary fields:\n",
+ "{}\n"
+ ]
+ }
+ ],
+ "source": [
+ "trainer.create_model_card(**kwargs)"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 32,
+ "id": "5c8abdc9",
+ "metadata": {},
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "Configuration saved in vitouphy/wav2vec2-xls-r-1b-km/config.json\n",
+ "Model weights saved in vitouphy/wav2vec2-xls-r-1b-km/pytorch_model.bin\n"
+ ]
+ },
+ {
+ "data": {
+ "application/vnd.jupyter.widget-view+json": {
+ "model_id": "2b7d33e02e974830824e8c1acccc1dbf",
+ "version_major": 2,
+ "version_minor": 0
+ },
+ "text/plain": [
+ "Upload file pytorch_model.bin: 0%| | 3.38k/3.59G [00:00, ?B/s]"
+ ]
+ },
+ "metadata": {},
+ "output_type": "display_data"
+ },
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "To https://huggingface.co/vitouphy/wav2vec2-xls-r-1b-km\n",
+ " 953f61b..183c558 main -> main\n",
+ "\n"
+ ]
+ },
+ {
+ "data": {
+ "text/plain": [
+ "'https://huggingface.co/vitouphy/wav2vec2-xls-r-1b-km/commit/183c558dd0aa166ae36561174695557ff4117eed'"
+ ]
+ },
+ "execution_count": 32,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "# model.push_to_hub('vitouphy/wav2vec2-xls-r-1b-km')"
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "id": "7dfead73",
+ "metadata": {},
+ "outputs": [],
+ "source": []
+ }
+ ],
+ "metadata": {
+ "kernelspec": {
+ "display_name": "Python 3 (ipykernel)",
+ "language": "python",
+ "name": "python3"
+ },
+ "language_info": {
+ "codemirror_mode": {
+ "name": "ipython",
+ "version": 3
+ },
+ "file_extension": ".py",
+ "mimetype": "text/x-python",
+ "name": "python",
+ "nbconvert_exporter": "python",
+ "pygments_lexer": "ipython3",
+ "version": "3.8.8"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}