src/test_whisper.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9d7b03aae28b4282b143eb17c3d8d687",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from huggingface_hub import notebook_login\n",
    "\n",
    "notebook_login()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Found cached dataset vivos (/home/tesla/.cache/huggingface/datasets/vivos/default/1.1.0/ab59078eb266c1a0ea856786ba56b5b8d56f29b42dfb37d92115cf81a7b1a5e0)\n",
      "Found cached dataset vivos (/home/tesla/.cache/huggingface/datasets/vivos/default/1.1.0/ab59078eb266c1a0ea856786ba56b5b8d56f29b42dfb37d92115cf81a7b1a5e0)\n"
     ]
    }
   ],
   "source": [
    "from datasets import load_dataset, DatasetDict\n",
    "\n",
    "vivos = DatasetDict()\n",
    "\n",
    "vivos[\"train\"] = load_dataset(\"vivos\", split=\"train\", use_auth_token=True)\n",
    "vivos[\"test\"] = load_dataset(\"vivos\", split=\"test\", use_auth_token=True)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['speaker_id', 'path', 'audio', 'sentence'],\n",
       "        num_rows: 11660\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['speaker_id', 'path', 'audio', 'sentence'],\n",
       "        num_rows: 760\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vivos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "vivos_clean = vivos.remove_columns([\"speaker_id\", \"path\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['audio', 'sentence'],\n",
       "        num_rows: 11660\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['audio', 'sentence'],\n",
       "        num_rows: 760\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vivos_clean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'audio': {'path': 'vivos/train/waves/VIVOSSPK27/VIVOSSPK27_084.wav', 'array': array([ 0.00000000e+00,  0.00000000e+00,  0.00000000e+00, ...,\n",
      "        9.15527344e-05, -5.18798828e-04, -9.15527344e-04]), 'sampling_rate': 16000}, 'sentence': 'CHƯA HẾT ĐI KHIẾU NẠI THÌ NHÀ MẠNG BẢO VỀ ĐẠI LÝ CHỌN SỐ KHÁC ĐI'}\n"
     ]
    }
   ],
   "source": [
    "print(vivos_clean['train'][12])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Found cached dataset common_voice_13_0 (/home/tesla/.cache/huggingface/datasets/mozilla-foundation___common_voice_13_0/vi/13.0.0/2506e9a8950f5807ceae08c2920e814222909fd7f477b74f5d225802e9f04055)\n",
      "Found cached dataset common_voice_13_0 (/home/tesla/.cache/huggingface/datasets/mozilla-foundation___common_voice_13_0/vi/13.0.0/2506e9a8950f5807ceae08c2920e814222909fd7f477b74f5d225802e9f04055)\n"
     ]
    }
   ],
   "source": [
    "\n",
    "common_voice = DatasetDict()\n",
    "\n",
    "common_voice[\"train\"] = load_dataset(\"mozilla-foundation/common_voice_13_0\", \"vi\", split=\"train+validation\", use_auth_token=True)\n",
    "common_voice[\"test\"] = load_dataset(\"mozilla-foundation/common_voice_13_0\", \"vi\", split=\"test\", use_auth_token=True)\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['client_id', 'path', 'audio', 'sentence', 'up_votes', 'down_votes', 'age', 'gender', 'accent', 'locale', 'segment', 'variant'],\n",
       "        num_rows: 2854\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['client_id', 'path', 'audio', 'sentence', 'up_votes', 'down_votes', 'age', 'gender', 'accent', 'locale', 'segment', 'variant'],\n",
       "        num_rows: 1225\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "common_voice"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "common_voice_clean = common_voice.remove_columns([\"client_id\", \"path\", \"down_votes\", \"gender\", \"locale\", \"segment\", \"up_votes\", \"age\", \"accent\", \"variant\"])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['audio', 'sentence'],\n",
       "        num_rows: 2854\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['audio', 'sentence'],\n",
       "        num_rows: 1225\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "common_voice_clean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'common_voice_clean' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[1], line 9\u001b[0m\n\u001b[1;32m      5\u001b[0m     \u001b[39mreturn\u001b[39;00m example\n\u001b[1;32m      7\u001b[0m common_voice_clear \u001b[39m=\u001b[39m DatasetDict()\n\u001b[0;32m----> 9\u001b[0m common_voice_clear[\u001b[39m\"\u001b[39m\u001b[39mtrain\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m common_voice_clean[\u001b[39m\"\u001b[39m\u001b[39mtrain\u001b[39m\u001b[39m\"\u001b[39m]\u001b[39m.\u001b[39mmap(convert_to_uppercase)\n\u001b[1;32m     10\u001b[0m common_voice_clear[\u001b[39m\"\u001b[39m\u001b[39mtest\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m common_voice_clean[\u001b[39m\"\u001b[39m\u001b[39mtest\u001b[39m\u001b[39m\"\u001b[39m]\u001b[39m.\u001b[39mmap(convert_to_uppercase)\n",
      "\u001b[0;31mNameError\u001b[0m: name 'common_voice_clean' is not defined"
     ]
    }
   ],
   "source": [
    "from datasets import DatasetDict\n",
    "\n",
    "def convert_to_uppercase(example):\n",
    "    example[\"sentence\"] = example[\"sentence\"].upper()\n",
    "    return example\n",
    "\n",
    "common_voice_clear = DatasetDict()\n",
    "\n",
    "common_voice_clear[\"train\"] = common_voice_clean[\"train\"].map(convert_to_uppercase)\n",
    "common_voice_clear[\"test\"] = common_voice_clean[\"test\"].map(convert_to_uppercase)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'audio': {'path': '/home/tesla/.cache/huggingface/datasets/downloads/extracted/acb70896120347904e003bb826dcabc1ddd05a02210935cb44ce1c807e8742a5/vi_train_0/common_voice_vi_23901118.mp3', 'array': array([ 0.00000000e+00,  4.20543185e-14,  1.38823347e-14, ...,\n",
      "       -8.41874498e-06, -8.36193431e-06, -6.76584477e-06]), 'sampling_rate': 48000}, 'sentence': 'KHI CON CÓ MẸ'}\n"
     ]
    }
   ],
   "source": [
    "print(common_voice_clear['train'][1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 2854/2854 [33:25<00:00,  1.42it/s]\n"
     ]
    }
   ],
   "source": [
    "from pydub import AudioSegment\n",
    "import os\n",
    "\n",
    "from tqdm import tqdm\n",
    "\n",
    "def convert_mp3_to_wav(mp3_path, wav_path, target_sampling_rate):\n",
    "    audio = AudioSegment.from_mp3(mp3_path)\n",
    "    audio = audio.set_frame_rate(target_sampling_rate)\n",
    "    audio.export(wav_path, format='wav')\n",
    "\n",
    "target_sampling_rate = 16000\n",
    "\n",
    "for example in tqdm(common_voice_clear[\"train\"]):\n",
    "    mp3_path = example[\"audio\"][\"path\"]\n",
    "    wav_path = os.path.splitext(mp3_path)[0] + \".wav\"\n",
    "    convert_mp3_to_wav(mp3_path, wav_path, target_sampling_rate)\n",
    "    example[\"audio\"][\"path\"] = wav_path\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "import datasets\n",
    "from datasets import Audio\n",
    "\n",
    "common_voice_clean = common_voice_clean.cast_column(\"audio\", Audio(sampling_rate=16000))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "concat = DatasetDict()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "concat[\"train\"] = datasets.concatenate_datasets([common_voice_clean[\"train\"], vivos_clean[\"train\"]])\n",
    "\n",
    "#concat['test']= datasets.concatenate_datasets([common_voice_clean[\"test\"], vivos_clean[\"test\"]])\n",
    "concat['test']= vivos_clean[\"test\"]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['audio', 'sentence'],\n",
       "        num_rows: 14514\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['audio', 'sentence'],\n",
       "        num_rows: 760\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "concat"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import WhisperFeatureExtractor\n",
    "\n",
    "feature_extractor = WhisperFeatureExtractor.from_pretrained(\"openai/whisper-small\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import WhisperTokenizerFast\n",
    "\n",
    "tokenizer = WhisperTokenizerFast.from_pretrained(\"openai/whisper-small\", language=\"Vietnamese\", task=\"transcribe\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input:                 KHÔNG CÓ AI BÁC BỎ QUYỀN ĐÓ\n",
      "Decoded w/ special:    <|startoftranscript|><|notimestamps|>KHÔNG CÓ AI BÁC BỎ QUYỀN ĐÓ<|endoftext|>\n",
      "Decoded w/out special: KHÔNG CÓ AI BÁC BỎ QUYỀN ĐÓ\n",
      "Are equal:             True\n"
     ]
    }
   ],
   "source": [
    "input_str = concat[\"train\"][8550][\"sentence\"]\n",
    "labels = tokenizer(input_str).input_ids\n",
    "decoded_with_special = tokenizer.decode(labels, skip_special_tokens=False)\n",
    "decoded_str = tokenizer.decode(labels, skip_special_tokens=True)\n",
    "\n",
    "print(f\"Input:                 {input_str}\")\n",
    "print(f\"Decoded w/ special:    {decoded_with_special}\")\n",
    "print(f\"Decoded w/out special: {decoded_str}\")\n",
    "print(f\"Are equal:             {input_str == decoded_str}\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import WhisperProcessor\n",
    "\n",
    "processor = WhisperProcessor.from_pretrained(\"openai/whisper-small\", language=\"Vietnamese\", task=\"transcribe\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import Audio\n",
    "\n",
    "concat = concat.cast_column(\"audio\", Audio(sampling_rate=16000))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'audio': {'path': 'vivos/train/waves/VIVOSSPK12/VIVOSSPK12_R077.wav', 'array': array([ 0.00000000e+00,  0.00000000e+00, -3.05175781e-05, ...,\n",
      "        1.31225586e-03,  1.12915039e-03,  1.55639648e-03]), 'sampling_rate': 16000}, 'sentence': 'KIÊN GIANG'}\n"
     ]
    }
   ],
   "source": [
    "print(concat[\"train\"][4500])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [],
   "source": [
    "def prepare_dataset(batch):\n",
    "    # load and resample audio data from 48 to 16kHz\n",
    "    audio = batch[\"audio\"]\n",
    "\n",
    "    # compute log-Mel input features from input audio array \n",
    "    batch[\"input_features\"] = feature_extractor(audio[\"array\"], sampling_rate=audio[\"sampling_rate\"]).input_features[0]\n",
    "\n",
    "    # encode target text to label ids \n",
    "    batch[\"labels\"] = tokenizer(batch[\"sentence\"]).input_ids\n",
    "    return batch\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c35c921e0dde433fb0ef9346310238a3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map (num_proc=6):   0%|          | 0/14514 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8c5af4ed5f8141d2b0673972f7616941",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map (num_proc=6):   0%|          | 0/760 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "concat = concat.map(prepare_dataset, remove_columns=concat.column_names[\"train\"], num_proc=6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "\n",
    "from dataclasses import dataclass\n",
    "from typing import Any, Dict, List, Union\n",
    "\n",
    "@dataclass\n",
    "class DataCollatorSpeechSeq2SeqWithPadding:\n",
    "    processor: Any\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 lengths and need different padding methods\n",
    "        # first treat the audio inputs by simply returning torch tensors\n",
    "        input_features = [{\"input_features\": feature[\"input_features\"]} for feature in features]\n",
    "        batch = self.processor.feature_extractor.pad(input_features, return_tensors=\"pt\")\n",
    "\n",
    "        # get the tokenized label sequences\n",
    "        label_features = [{\"input_ids\": feature[\"labels\"]} for feature in features]\n",
    "        # pad the labels to max length\n",
    "        labels_batch = self.processor.tokenizer.pad(label_features, return_tensors=\"pt\")\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",
    "        # if bos token is appended in previous tokenization step,\n",
    "        # cut bos token here as it's append later anyways\n",
    "        if (labels[:, 0] == self.processor.tokenizer.bos_token_id).all().cpu().item():\n",
    "            labels = labels[:, 1:]\n",
    "\n",
    "        batch[\"labels\"] = labels\n",
    "\n",
    "        return batch\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_collator = DataCollatorSpeechSeq2SeqWithPadding(processor=processor)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Train\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [],
   "source": [
    "import evaluate\n",
    "\n",
    "metric = evaluate.load(\"wer\")\n",
    "\n",
    "\n",
    "def compute_metrics(pred):\n",
    "    pred_ids = pred.predictions\n",
    "    label_ids = pred.label_ids\n",
    "\n",
    "    # replace -100 with the pad_token_id\n",
    "    label_ids[label_ids == -100] = tokenizer.pad_token_id\n",
    "\n",
    "    # we do not want to group tokens when computing the metrics\n",
    "    pred_str = tokenizer.batch_decode(pred_ids, skip_special_tokens=True)\n",
    "    label_str = tokenizer.batch_decode(label_ids, skip_special_tokens=True)\n",
    "\n",
    "    wer = 100 * metric.compute(predictions=pred_str, references=label_str)\n",
    "\n",
    "    return {\"wer\": wer}\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import WhisperForConditionalGeneration\n",
    "\n",
    "model = WhisperForConditionalGeneration.from_pretrained(\"openai/whisper-small\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.config.forced_decoder_ids = None\n",
    "model.config.suppress_tokens = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import Seq2SeqTrainingArguments\n",
    "\n",
    "training_args = Seq2SeqTrainingArguments(\n",
    "    output_dir=\"./vi_whisper-small\",  # change to a repo name of your choice\n",
    "    per_device_train_batch_size=16,\n",
    "    gradient_accumulation_steps=1,  # increase by 2x for every 2x decrease in batch size\n",
    "    learning_rate=1e-4,\n",
    "    warmup_steps=1000,\n",
    "    max_steps=8000,\n",
    "    gradient_checkpointing=True,\n",
    "    fp16=True,\n",
    "    evaluation_strategy=\"steps\",\n",
    "    per_device_eval_batch_size=8,\n",
    "    predict_with_generate=True,\n",
    "    generation_max_length=225,\n",
    "    save_steps=4000,\n",
    "    eval_steps=1000,\n",
    "    logging_steps=25,\n",
    "    report_to=[\"tensorboard\"],\n",
    "    load_best_model_at_end=True,\n",
    "    metric_for_best_model=\"wer\",\n",
    "    greater_is_better=False,\n",
    "    push_to_hub=True,\n",
    ")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/media/tesla/New Volume1/DEMO/DUY/Vietnamese_ASR/./vi_whisper-small is already a clone of https://huggingface.co/DuyTa/vi_whisper-small. Make sure you pull the latest changes with `repo.git_pull()`.\n"
     ]
    },
    {
     "ename": "OSError",
     "evalue": "From https://huggingface.co/DuyTa/vi_whisper-small\n   d7893fc..47c00b5  main       -> origin/main\nhint: You have divergent branches and need to specify how to reconcile them.\nhint: You can do so by running one of the following commands sometime before\nhint: your next pull:\nhint: \nhint:   git config pull.rebase false  # merge (the default strategy)\nhint:   git config pull.rebase true   # rebase\nhint:   git config pull.ff only       # fast-forward only\nhint: \nhint: You can replace \"git config\" with \"git config --global\" to set a default\nhint: preference for all repositories. You can also pass --rebase, --no-rebase,\nhint: or --ff-only on the command line to override the configured default per\nhint: invocation.\nfatal: Need to specify how to reconcile divergent branches.\n",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mCalledProcessError\u001b[0m                        Traceback (most recent call last)",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/repository.py:984\u001b[0m, in \u001b[0;36mRepository.git_pull\u001b[0;34m(self, rebase, lfs)\u001b[0m\n\u001b[1;32m    983\u001b[0m \u001b[39mwith\u001b[39;00m _lfs_log_progress():\n\u001b[0;32m--> 984\u001b[0m     result \u001b[39m=\u001b[39m run_subprocess(command, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mlocal_dir)\n\u001b[1;32m    985\u001b[0m     logger\u001b[39m.\u001b[39minfo(result\u001b[39m.\u001b[39mstdout)\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/utils/_subprocess.py:83\u001b[0m, in \u001b[0;36mrun_subprocess\u001b[0;34m(command, folder, check, **kwargs)\u001b[0m\n\u001b[1;32m     81\u001b[0m     folder \u001b[39m=\u001b[39m \u001b[39mstr\u001b[39m(folder)\n\u001b[0;32m---> 83\u001b[0m \u001b[39mreturn\u001b[39;00m subprocess\u001b[39m.\u001b[39;49mrun(\n\u001b[1;32m     84\u001b[0m     command,\n\u001b[1;32m     85\u001b[0m     stderr\u001b[39m=\u001b[39;49msubprocess\u001b[39m.\u001b[39;49mPIPE,\n\u001b[1;32m     86\u001b[0m     stdout\u001b[39m=\u001b[39;49msubprocess\u001b[39m.\u001b[39;49mPIPE,\n\u001b[1;32m     87\u001b[0m     check\u001b[39m=\u001b[39;49mcheck,\n\u001b[1;32m     88\u001b[0m     encoding\u001b[39m=\u001b[39;49m\u001b[39m\"\u001b[39;49m\u001b[39mutf-8\u001b[39;49m\u001b[39m\"\u001b[39;49m,\n\u001b[1;32m     89\u001b[0m     errors\u001b[39m=\u001b[39;49m\u001b[39m\"\u001b[39;49m\u001b[39mreplace\u001b[39;49m\u001b[39m\"\u001b[39;49m,  \u001b[39m# if not utf-8, replace char by �\u001b[39;49;00m\n\u001b[1;32m     90\u001b[0m     cwd\u001b[39m=\u001b[39;49mfolder \u001b[39mor\u001b[39;49;00m os\u001b[39m.\u001b[39;49mgetcwd(),\n\u001b[1;32m     91\u001b[0m     \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs,\n\u001b[1;32m     92\u001b[0m )\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/subprocess.py:528\u001b[0m, in \u001b[0;36mrun\u001b[0;34m(input, capture_output, timeout, check, *popenargs, **kwargs)\u001b[0m\n\u001b[1;32m    527\u001b[0m     \u001b[39mif\u001b[39;00m check \u001b[39mand\u001b[39;00m retcode:\n\u001b[0;32m--> 528\u001b[0m         \u001b[39mraise\u001b[39;00m CalledProcessError(retcode, process\u001b[39m.\u001b[39margs,\n\u001b[1;32m    529\u001b[0m                                  output\u001b[39m=\u001b[39mstdout, stderr\u001b[39m=\u001b[39mstderr)\n\u001b[1;32m    530\u001b[0m \u001b[39mreturn\u001b[39;00m CompletedProcess(process\u001b[39m.\u001b[39margs, retcode, stdout, stderr)\n",
      "\u001b[0;31mCalledProcessError\u001b[0m: Command '['git', 'pull']' returned non-zero exit status 128.",
      "\nDuring handling of the above exception, another exception occurred:\n",
      "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[126], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mtransformers\u001b[39;00m \u001b[39mimport\u001b[39;00m Seq2SeqTrainer\n\u001b[0;32m----> 3\u001b[0m trainer \u001b[39m=\u001b[39m Seq2SeqTrainer(\n\u001b[1;32m      4\u001b[0m     args\u001b[39m=\u001b[39;49mtraining_args,\n\u001b[1;32m      5\u001b[0m     model\u001b[39m=\u001b[39;49mmodel,\n\u001b[1;32m      6\u001b[0m     train_dataset\u001b[39m=\u001b[39;49mconcat[\u001b[39m\"\u001b[39;49m\u001b[39mtrain\u001b[39;49m\u001b[39m\"\u001b[39;49m],\n\u001b[1;32m      7\u001b[0m \n\u001b[1;32m      8\u001b[0m \n\u001b[1;32m      9\u001b[0m \n\u001b[1;32m     10\u001b[0m \n\u001b[1;32m     11\u001b[0m     eval_dataset\u001b[39m=\u001b[39;49mconcat[\u001b[39m\"\u001b[39;49m\u001b[39mtest\u001b[39;49m\u001b[39m\"\u001b[39;49m],\n\u001b[1;32m     12\u001b[0m     data_collator\u001b[39m=\u001b[39;49mdata_collator,\n\u001b[1;32m     13\u001b[0m     compute_metrics\u001b[39m=\u001b[39;49mcompute_metrics,\n\u001b[1;32m     14\u001b[0m     tokenizer\u001b[39m=\u001b[39;49mprocessor\u001b[39m.\u001b[39;49mfeature_extractor,\n\u001b[1;32m     15\u001b[0m )\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/trainer_seq2seq.py:56\u001b[0m, in \u001b[0;36mSeq2SeqTrainer.__init__\u001b[0;34m(self, model, args, data_collator, train_dataset, eval_dataset, tokenizer, model_init, compute_metrics, callbacks, optimizers, preprocess_logits_for_metrics)\u001b[0m\n\u001b[1;32m     42\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39m__init__\u001b[39m(\n\u001b[1;32m     43\u001b[0m     \u001b[39mself\u001b[39m,\n\u001b[1;32m     44\u001b[0m     model: Union[\u001b[39m\"\u001b[39m\u001b[39mPreTrainedModel\u001b[39m\u001b[39m\"\u001b[39m, nn\u001b[39m.\u001b[39mModule] \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m     54\u001b[0m     preprocess_logits_for_metrics: Optional[Callable[[torch\u001b[39m.\u001b[39mTensor, torch\u001b[39m.\u001b[39mTensor], torch\u001b[39m.\u001b[39mTensor]] \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m,\n\u001b[1;32m     55\u001b[0m ):\n\u001b[0;32m---> 56\u001b[0m     \u001b[39msuper\u001b[39;49m()\u001b[39m.\u001b[39;49m\u001b[39m__init__\u001b[39;49m(\n\u001b[1;32m     57\u001b[0m         model\u001b[39m=\u001b[39;49mmodel,\n\u001b[1;32m     58\u001b[0m         args\u001b[39m=\u001b[39;49margs,\n\u001b[1;32m     59\u001b[0m         data_collator\u001b[39m=\u001b[39;49mdata_collator,\n\u001b[1;32m     60\u001b[0m         train_dataset\u001b[39m=\u001b[39;49mtrain_dataset,\n\u001b[1;32m     61\u001b[0m         eval_dataset\u001b[39m=\u001b[39;49meval_dataset,\n\u001b[1;32m     62\u001b[0m         tokenizer\u001b[39m=\u001b[39;49mtokenizer,\n\u001b[1;32m     63\u001b[0m         model_init\u001b[39m=\u001b[39;49mmodel_init,\n\u001b[1;32m     64\u001b[0m         compute_metrics\u001b[39m=\u001b[39;49mcompute_metrics,\n\u001b[1;32m     65\u001b[0m         callbacks\u001b[39m=\u001b[39;49mcallbacks,\n\u001b[1;32m     66\u001b[0m         optimizers\u001b[39m=\u001b[39;49moptimizers,\n\u001b[1;32m     67\u001b[0m         preprocess_logits_for_metrics\u001b[39m=\u001b[39;49mpreprocess_logits_for_metrics,\n\u001b[1;32m     68\u001b[0m     )\n\u001b[1;32m     70\u001b[0m     \u001b[39m# Override self.model.generation_config if a GenerationConfig is specified in args.\u001b[39;00m\n\u001b[1;32m     71\u001b[0m     \u001b[39m# Priority: args.generation_config > model.generation_config > default GenerationConfig.\u001b[39;00m\n\u001b[1;32m     72\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39margs\u001b[39m.\u001b[39mgeneration_config \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/trainer.py:551\u001b[0m, in \u001b[0;36mTrainer.__init__\u001b[0;34m(self, model, args, data_collator, train_dataset, eval_dataset, tokenizer, model_init, compute_metrics, callbacks, optimizers, preprocess_logits_for_metrics)\u001b[0m\n\u001b[1;32m    549\u001b[0m \u001b[39m# Create clone of distant repo and output directory if needed\u001b[39;00m\n\u001b[1;32m    550\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39margs\u001b[39m.\u001b[39mpush_to_hub:\n\u001b[0;32m--> 551\u001b[0m     \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49minit_git_repo(at_init\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m)\n\u001b[1;32m    552\u001b[0m     \u001b[39m# In case of pull, we need to make sure every process has the latest.\u001b[39;00m\n\u001b[1;32m    553\u001b[0m     \u001b[39mif\u001b[39;00m is_torch_tpu_available():\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/trainer.py:3449\u001b[0m, in \u001b[0;36mTrainer.init_git_repo\u001b[0;34m(self, at_init)\u001b[0m\n\u001b[1;32m   3446\u001b[0m     \u001b[39melse\u001b[39;00m:\n\u001b[1;32m   3447\u001b[0m         \u001b[39mraise\u001b[39;00m\n\u001b[0;32m-> 3449\u001b[0m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mrepo\u001b[39m.\u001b[39;49mgit_pull()\n\u001b[1;32m   3451\u001b[0m \u001b[39m# By default, ignore the checkpoint folders\u001b[39;00m\n\u001b[1;32m   3452\u001b[0m \u001b[39mif\u001b[39;00m (\n\u001b[1;32m   3453\u001b[0m     \u001b[39mnot\u001b[39;00m os\u001b[39m.\u001b[39mpath\u001b[39m.\u001b[39mexists(os\u001b[39m.\u001b[39mpath\u001b[39m.\u001b[39mjoin(\u001b[39mself\u001b[39m\u001b[39m.\u001b[39margs\u001b[39m.\u001b[39moutput_dir, \u001b[39m\"\u001b[39m\u001b[39m.gitignore\u001b[39m\u001b[39m\"\u001b[39m))\n\u001b[1;32m   3454\u001b[0m     \u001b[39mand\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39margs\u001b[39m.\u001b[39mhub_strategy \u001b[39m!=\u001b[39m HubStrategy\u001b[39m.\u001b[39mALL_CHECKPOINTS\n\u001b[1;32m   3455\u001b[0m ):\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/repository.py:987\u001b[0m, in \u001b[0;36mRepository.git_pull\u001b[0;34m(self, rebase, lfs)\u001b[0m\n\u001b[1;32m    985\u001b[0m         logger\u001b[39m.\u001b[39minfo(result\u001b[39m.\u001b[39mstdout)\n\u001b[1;32m    986\u001b[0m \u001b[39mexcept\u001b[39;00m subprocess\u001b[39m.\u001b[39mCalledProcessError \u001b[39mas\u001b[39;00m exc:\n\u001b[0;32m--> 987\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mEnvironmentError\u001b[39;00m(exc\u001b[39m.\u001b[39mstderr)\n",
      "\u001b[0;31mOSError\u001b[0m: From https://huggingface.co/DuyTa/vi_whisper-small\n   d7893fc..47c00b5  main       -> origin/main\nhint: You have divergent branches and need to specify how to reconcile them.\nhint: You can do so by running one of the following commands sometime before\nhint: your next pull:\nhint: \nhint:   git config pull.rebase false  # merge (the default strategy)\nhint:   git config pull.rebase true   # rebase\nhint:   git config pull.ff only       # fast-forward only\nhint: \nhint: You can replace \"git config\" with \"git config --global\" to set a default\nhint: preference for all repositories. You can also pass --rebase, --no-rebase,\nhint: or --ff-only on the command line to override the configured default per\nhint: invocation.\nfatal: Need to specify how to reconcile divergent branches.\n"
     ]
    }
   ],
   "source": [
    "from transformers import Seq2SeqTrainer\n",
    "\n",
    "trainer = Seq2SeqTrainer(\n",
    "    args=training_args,\n",
    "    model=model,\n",
    "    train_dataset=concat[\"train\"],\n",
    "\n",
    "    eval_dataset=concat[\"test\"],\n",
    "    data_collator=data_collator,\n",
    "    compute_metrics=compute_metrics,\n",
    "    tokenizer=processor.feature_extractor,\n",
    ")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('./vi_whisper-small/tokenizer_config.json',\n",
       " './vi_whisper-small/special_tokens_map.json',\n",
       " './vi_whisper-small/vocab.json',\n",
       " './vi_whisper-small/merges.txt',\n",
       " './vi_whisper-small/normalizer.json',\n",
       " './vi_whisper-small/added_tokens.json',\n",
       " './vi_whisper-small/tokenizer.json')"
      ]
     },
     "execution_count": 130,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.save_pretrained(\"./vi_whisper-small/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Device 0:\n",
      "  Currently allocated memory: 922.884765625 MB\n",
      "  Peak memory usage: 922.884765625 MB\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "\n",
    "device_count = torch.cuda.device_count()\n",
    "\n",
    "for device in range(device_count):\n",
    "    torch.cuda.device(device)\n",
    "    allocated_memory = torch.cuda.memory_allocated(device)\n",
    "    peak_memory = torch.cuda.max_memory_allocated(device)\n",
    "    print(f\"Device {device}:\")\n",
    "    print(f\"  Currently allocated memory: {allocated_memory / 1024**2} MB\")\n",
    "    print(f\"  Peak memory usage: {peak_memory / 1024**2} MB\")\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Device 0:\n",
      "  Name: Tesla T4\n",
      "  Max Memory: 14966.375 MB\n"
     ]
    }
   ],
   "source": [
    "device_count = torch.cuda.device_count()\n",
    "\n",
    "for device in range(device_count):\n",
    "    properties = torch.cuda.get_device_properties(device)\n",
    "    print(f\"Device {device}:\")\n",
    "    print(f\"  Name: {properties.name}\")\n",
    "    print(f\"  Max Memory: {properties.total_memory / 1024**2} MB\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 111,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/tesla/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/optimization.py:411: FutureWarning: This implementation of AdamW is deprecated and will be removed in a future version. Use the PyTorch implementation torch.optim.AdamW instead, or set `no_deprecation_warning=True` to disable this warning\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "da5a536a979e4d34bc59eaede9204d06",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/8000 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'loss': 3.8537, 'learning_rate': 2.1000000000000002e-06, 'epoch': 0.03}\n",
      "{'loss': 2.2347, 'learning_rate': 4.6e-06, 'epoch': 0.06}\n",
      "{'loss': 1.2627, 'learning_rate': 7.1e-06, 'epoch': 0.08}\n",
      "{'loss': 0.8976, 'learning_rate': 9.600000000000001e-06, 'epoch': 0.11}\n",
      "{'loss': 0.7313, 'learning_rate': 1.2100000000000001e-05, 'epoch': 0.14}\n",
      "{'loss': 0.6526, 'learning_rate': 1.4599999999999999e-05, 'epoch': 0.17}\n",
      "{'loss': 0.7221, 'learning_rate': 1.7000000000000003e-05, 'epoch': 0.19}\n",
      "{'loss': 0.6478, 'learning_rate': 1.9500000000000003e-05, 'epoch': 0.22}\n",
      "{'loss': 1.7029, 'learning_rate': 2.19e-05, 'epoch': 0.25}\n",
      "{'loss': 1.1476, 'learning_rate': 2.44e-05, 'epoch': 0.28}\n",
      "{'loss': 0.5837, 'learning_rate': 2.6900000000000003e-05, 'epoch': 0.3}\n",
      "{'loss': 0.5912, 'learning_rate': 2.94e-05, 'epoch': 0.33}\n",
      "{'loss': 0.6872, 'learning_rate': 3.19e-05, 'epoch': 0.36}\n",
      "{'loss': 0.4103, 'learning_rate': 3.4399999999999996e-05, 'epoch': 0.39}\n",
      "{'loss': 0.4293, 'learning_rate': 3.69e-05, 'epoch': 0.41}\n",
      "{'loss': 0.3055, 'learning_rate': 3.94e-05, 'epoch': 0.44}\n",
      "{'loss': 0.311, 'learning_rate': 4.19e-05, 'epoch': 0.47}\n",
      "{'loss': 0.3212, 'learning_rate': 4.44e-05, 'epoch': 0.5}\n",
      "{'loss': 0.2917, 'learning_rate': 4.69e-05, 'epoch': 0.52}\n",
      "{'loss': 0.2975, 'learning_rate': 4.94e-05, 'epoch': 0.55}\n",
      "{'loss': 0.3254, 'learning_rate': 5.19e-05, 'epoch': 0.58}\n",
      "{'loss': 0.2825, 'learning_rate': 5.440000000000001e-05, 'epoch': 0.61}\n",
      "{'loss': 0.2929, 'learning_rate': 5.69e-05, 'epoch': 0.63}\n",
      "{'loss': 0.3056, 'learning_rate': 5.94e-05, 'epoch': 0.66}\n",
      "{'loss': 0.3105, 'learning_rate': 6.19e-05, 'epoch': 0.69}\n",
      "{'loss': 0.3702, 'learning_rate': 6.440000000000001e-05, 'epoch': 0.72}\n",
      "{'loss': 0.2684, 'learning_rate': 6.690000000000001e-05, 'epoch': 0.74}\n",
      "{'loss': 0.2767, 'learning_rate': 6.939999999999999e-05, 'epoch': 0.77}\n",
      "{'loss': 0.315, 'learning_rate': 7.19e-05, 'epoch': 0.8}\n",
      "{'loss': 0.3132, 'learning_rate': 7.44e-05, 'epoch': 0.83}\n",
      "{'loss': 0.3933, 'learning_rate': 7.69e-05, 'epoch': 0.85}\n",
      "{'loss': 0.311, 'learning_rate': 7.94e-05, 'epoch': 0.88}\n",
      "{'loss': 0.3104, 'learning_rate': 8.19e-05, 'epoch': 0.91}\n",
      "{'loss': 0.297, 'learning_rate': 8.44e-05, 'epoch': 0.94}\n",
      "{'loss': 0.3094, 'learning_rate': 8.69e-05, 'epoch': 0.96}\n",
      "{'loss': 0.29, 'learning_rate': 8.94e-05, 'epoch': 0.99}\n",
      "{'loss': 0.2712, 'learning_rate': 9.190000000000001e-05, 'epoch': 1.02}\n",
      "{'loss': 0.262, 'learning_rate': 9.44e-05, 'epoch': 1.05}\n",
      "{'loss': 0.2481, 'learning_rate': 9.69e-05, 'epoch': 1.07}\n",
      "{'loss': 0.249, 'learning_rate': 9.94e-05, 'epoch': 1.1}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ea1e7cf193cc4b5dacd0883200ff6ef6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.3765707015991211, 'eval_wer': 32.16783216783217, 'eval_runtime': 349.1035, 'eval_samples_per_second': 2.177, 'eval_steps_per_second': 0.272, 'epoch': 1.1}\n",
      "{'loss': 0.2729, 'learning_rate': 9.972857142857144e-05, 'epoch': 1.13}\n",
      "{'loss': 0.267, 'learning_rate': 9.937142857142857e-05, 'epoch': 1.16}\n",
      "{'loss': 0.2617, 'learning_rate': 9.901428571428571e-05, 'epoch': 1.18}\n",
      "{'loss': 0.2613, 'learning_rate': 9.865714285714286e-05, 'epoch': 1.21}\n",
      "{'loss': 0.2736, 'learning_rate': 9.83e-05, 'epoch': 1.24}\n",
      "{'loss': 0.245, 'learning_rate': 9.794285714285714e-05, 'epoch': 1.27}\n",
      "{'loss': 0.2385, 'learning_rate': 9.75857142857143e-05, 'epoch': 1.29}\n",
      "{'loss': 0.258, 'learning_rate': 9.722857142857144e-05, 'epoch': 1.32}\n",
      "{'loss': 0.2623, 'learning_rate': 9.687142857142858e-05, 'epoch': 1.35}\n",
      "{'loss': 0.2346, 'learning_rate': 9.651428571428572e-05, 'epoch': 1.38}\n",
      "{'loss': 0.2376, 'learning_rate': 9.615714285714286e-05, 'epoch': 1.4}\n",
      "{'loss': 0.246, 'learning_rate': 9.58e-05, 'epoch': 1.43}\n",
      "{'loss': 0.2201, 'learning_rate': 9.544285714285715e-05, 'epoch': 1.46}\n",
      "{'loss': 0.2233, 'learning_rate': 9.508571428571429e-05, 'epoch': 1.49}\n",
      "{'loss': 0.2154, 'learning_rate': 9.472857142857143e-05, 'epoch': 1.51}\n",
      "{'loss': 0.2348, 'learning_rate': 9.437142857142857e-05, 'epoch': 1.54}\n",
      "{'loss': 0.2159, 'learning_rate': 9.401428571428572e-05, 'epoch': 1.57}\n",
      "{'loss': 0.2265, 'learning_rate': 9.365714285714286e-05, 'epoch': 1.6}\n",
      "{'loss': 0.2118, 'learning_rate': 9.33e-05, 'epoch': 1.62}\n",
      "{'loss': 0.2223, 'learning_rate': 9.294285714285714e-05, 'epoch': 1.65}\n",
      "{'loss': 0.2, 'learning_rate': 9.258571428571428e-05, 'epoch': 1.68}\n",
      "{'loss': 0.206, 'learning_rate': 9.222857142857142e-05, 'epoch': 1.71}\n",
      "{'loss': 0.1979, 'learning_rate': 9.187142857142858e-05, 'epoch': 1.73}\n",
      "{'loss': 0.2022, 'learning_rate': 9.151428571428572e-05, 'epoch': 1.76}\n",
      "{'loss': 0.2028, 'learning_rate': 9.115714285714286e-05, 'epoch': 1.79}\n",
      "{'loss': 0.2161, 'learning_rate': 9.080000000000001e-05, 'epoch': 1.82}\n",
      "{'loss': 0.1964, 'learning_rate': 9.044285714285715e-05, 'epoch': 1.84}\n",
      "{'loss': 0.2151, 'learning_rate': 9.008571428571429e-05, 'epoch': 1.87}\n",
      "{'loss': 0.2056, 'learning_rate': 8.972857142857143e-05, 'epoch': 1.9}\n",
      "{'loss': 0.189, 'learning_rate': 8.937142857142857e-05, 'epoch': 1.93}\n",
      "{'loss': 0.1944, 'learning_rate': 8.901428571428571e-05, 'epoch': 1.95}\n",
      "{'loss': 0.1834, 'learning_rate': 8.865714285714287e-05, 'epoch': 1.98}\n",
      "{'loss': 0.1557, 'learning_rate': 8.83e-05, 'epoch': 2.01}\n",
      "{'loss': 0.1337, 'learning_rate': 8.794285714285714e-05, 'epoch': 2.04}\n",
      "{'loss': 0.1338, 'learning_rate': 8.75857142857143e-05, 'epoch': 2.06}\n",
      "{'loss': 0.1338, 'learning_rate': 8.722857142857144e-05, 'epoch': 2.09}\n",
      "{'loss': 0.1385, 'learning_rate': 8.687142857142856e-05, 'epoch': 2.12}\n",
      "{'loss': 0.1259, 'learning_rate': 8.651428571428572e-05, 'epoch': 2.15}\n",
      "{'loss': 0.1268, 'learning_rate': 8.615714285714286e-05, 'epoch': 2.18}\n",
      "{'loss': 0.1416, 'learning_rate': 8.58e-05, 'epoch': 2.2}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2b48a513542d43558d8fef6a8ef00629",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.2880653738975525, 'eval_wer': 46.464646464646464, 'eval_runtime': 337.0754, 'eval_samples_per_second': 2.255, 'eval_steps_per_second': 0.282, 'epoch': 2.2}\n",
      "{'loss': 0.1271, 'learning_rate': 8.544285714285715e-05, 'epoch': 2.23}\n",
      "{'loss': 0.1345, 'learning_rate': 8.508571428571429e-05, 'epoch': 2.26}\n",
      "{'loss': 0.149, 'learning_rate': 8.472857142857143e-05, 'epoch': 2.29}\n",
      "{'loss': 0.1289, 'learning_rate': 8.437142857142859e-05, 'epoch': 2.31}\n",
      "{'loss': 0.1391, 'learning_rate': 8.401428571428573e-05, 'epoch': 2.34}\n",
      "{'loss': 0.1532, 'learning_rate': 8.365714285714285e-05, 'epoch': 2.37}\n",
      "{'loss': 0.1283, 'learning_rate': 8.33e-05, 'epoch': 2.4}\n",
      "{'loss': 0.1336, 'learning_rate': 8.294285714285715e-05, 'epoch': 2.42}\n",
      "{'loss': 0.129, 'learning_rate': 8.258571428571429e-05, 'epoch': 2.45}\n",
      "{'loss': 0.1399, 'learning_rate': 8.222857142857144e-05, 'epoch': 2.48}\n",
      "{'loss': 0.1411, 'learning_rate': 8.187142857142858e-05, 'epoch': 2.51}\n",
      "{'loss': 0.1298, 'learning_rate': 8.151428571428572e-05, 'epoch': 2.53}\n",
      "{'loss': 0.1397, 'learning_rate': 8.115714285714286e-05, 'epoch': 2.56}\n",
      "{'loss': 0.1356, 'learning_rate': 8.080000000000001e-05, 'epoch': 2.59}\n",
      "{'loss': 0.1366, 'learning_rate': 8.044285714285714e-05, 'epoch': 2.62}\n",
      "{'loss': 0.1331, 'learning_rate': 8.008571428571429e-05, 'epoch': 2.64}\n",
      "{'loss': 0.1297, 'learning_rate': 7.972857142857143e-05, 'epoch': 2.67}\n",
      "{'loss': 0.1414, 'learning_rate': 7.937142857142857e-05, 'epoch': 2.7}\n",
      "{'loss': 0.1189, 'learning_rate': 7.901428571428571e-05, 'epoch': 2.73}\n",
      "{'loss': 0.1416, 'learning_rate': 7.865714285714287e-05, 'epoch': 2.75}\n",
      "{'loss': 0.1378, 'learning_rate': 7.83e-05, 'epoch': 2.78}\n",
      "{'loss': 0.1305, 'learning_rate': 7.794285714285715e-05, 'epoch': 2.81}\n",
      "{'loss': 0.1571, 'learning_rate': 7.75857142857143e-05, 'epoch': 2.84}\n",
      "{'loss': 0.1285, 'learning_rate': 7.722857142857143e-05, 'epoch': 2.86}\n",
      "{'loss': 0.1339, 'learning_rate': 7.687142857142857e-05, 'epoch': 2.89}\n",
      "{'loss': 0.1216, 'learning_rate': 7.651428571428572e-05, 'epoch': 2.92}\n",
      "{'loss': 0.1321, 'learning_rate': 7.615714285714286e-05, 'epoch': 2.95}\n",
      "{'loss': 0.1259, 'learning_rate': 7.58e-05, 'epoch': 2.97}\n",
      "{'loss': 0.1259, 'learning_rate': 7.544285714285715e-05, 'epoch': 3.0}\n",
      "{'loss': 0.0851, 'learning_rate': 7.508571428571429e-05, 'epoch': 3.03}\n",
      "{'loss': 0.0764, 'learning_rate': 7.472857142857143e-05, 'epoch': 3.06}\n",
      "{'loss': 0.0986, 'learning_rate': 7.438571428571429e-05, 'epoch': 3.08}\n",
      "{'loss': 0.0883, 'learning_rate': 7.402857142857143e-05, 'epoch': 3.11}\n",
      "{'loss': 0.0811, 'learning_rate': 7.367142857142858e-05, 'epoch': 3.14}\n",
      "{'loss': 0.0872, 'learning_rate': 7.331428571428571e-05, 'epoch': 3.17}\n",
      "{'loss': 0.0872, 'learning_rate': 7.295714285714286e-05, 'epoch': 3.19}\n",
      "{'loss': 0.0805, 'learning_rate': 7.26e-05, 'epoch': 3.22}\n",
      "{'loss': 0.0803, 'learning_rate': 7.224285714285714e-05, 'epoch': 3.25}\n",
      "{'loss': 0.0753, 'learning_rate': 7.188571428571428e-05, 'epoch': 3.28}\n",
      "{'loss': 0.0839, 'learning_rate': 7.152857142857144e-05, 'epoch': 3.3}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ce95b9d2a270464fbeab22454f214a1d",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.279912531375885, 'eval_wer': 22.779072779072777, 'eval_runtime': 345.4945, 'eval_samples_per_second': 2.2, 'eval_steps_per_second': 0.275, 'epoch': 3.3}\n",
      "{'loss': 0.0885, 'learning_rate': 7.117142857142858e-05, 'epoch': 3.33}\n",
      "{'loss': 0.0845, 'learning_rate': 7.081428571428572e-05, 'epoch': 3.36}\n",
      "{'loss': 0.0761, 'learning_rate': 7.045714285714287e-05, 'epoch': 3.39}\n",
      "{'loss': 0.0756, 'learning_rate': 7.01e-05, 'epoch': 3.41}\n",
      "{'loss': 0.0859, 'learning_rate': 6.974285714285715e-05, 'epoch': 3.44}\n",
      "{'loss': 0.0972, 'learning_rate': 6.938571428571429e-05, 'epoch': 3.47}\n",
      "{'loss': 0.0822, 'learning_rate': 6.902857142857143e-05, 'epoch': 3.5}\n",
      "{'loss': 0.0892, 'learning_rate': 6.867142857142857e-05, 'epoch': 3.52}\n",
      "{'loss': 0.0735, 'learning_rate': 6.831428571428572e-05, 'epoch': 3.55}\n",
      "{'loss': 0.0893, 'learning_rate': 6.795714285714286e-05, 'epoch': 3.58}\n",
      "{'loss': 0.0869, 'learning_rate': 6.76e-05, 'epoch': 3.61}\n",
      "{'loss': 0.0877, 'learning_rate': 6.724285714285714e-05, 'epoch': 3.63}\n",
      "{'loss': 0.07, 'learning_rate': 6.688571428571428e-05, 'epoch': 3.66}\n",
      "{'loss': 0.0807, 'learning_rate': 6.652857142857142e-05, 'epoch': 3.69}\n",
      "{'loss': 0.0831, 'learning_rate': 6.617142857142858e-05, 'epoch': 3.72}\n",
      "{'loss': 0.0836, 'learning_rate': 6.581428571428572e-05, 'epoch': 3.74}\n",
      "{'loss': 0.0875, 'learning_rate': 6.545714285714286e-05, 'epoch': 3.77}\n",
      "{'loss': 0.0846, 'learning_rate': 6.510000000000001e-05, 'epoch': 3.8}\n",
      "{'loss': 0.0779, 'learning_rate': 6.474285714285715e-05, 'epoch': 3.83}\n",
      "{'loss': 0.0871, 'learning_rate': 6.438571428571429e-05, 'epoch': 3.85}\n",
      "{'loss': 0.0777, 'learning_rate': 6.402857142857143e-05, 'epoch': 3.88}\n",
      "{'loss': 0.0856, 'learning_rate': 6.367142857142857e-05, 'epoch': 3.91}\n",
      "{'loss': 0.083, 'learning_rate': 6.331428571428571e-05, 'epoch': 3.94}\n",
      "{'loss': 0.0667, 'learning_rate': 6.295714285714286e-05, 'epoch': 3.96}\n",
      "{'loss': 0.083, 'learning_rate': 6.26e-05, 'epoch': 3.99}\n",
      "{'loss': 0.0505, 'learning_rate': 6.224285714285714e-05, 'epoch': 4.02}\n",
      "{'loss': 0.0426, 'learning_rate': 6.18857142857143e-05, 'epoch': 4.05}\n",
      "{'loss': 0.0453, 'learning_rate': 6.152857142857144e-05, 'epoch': 4.07}\n",
      "{'loss': 0.0482, 'learning_rate': 6.117142857142858e-05, 'epoch': 4.1}\n",
      "{'loss': 0.0511, 'learning_rate': 6.081428571428571e-05, 'epoch': 4.13}\n",
      "{'loss': 0.0583, 'learning_rate': 6.045714285714286e-05, 'epoch': 4.16}\n",
      "{'loss': 0.0466, 'learning_rate': 6.0100000000000004e-05, 'epoch': 4.19}\n",
      "{'loss': 0.0502, 'learning_rate': 5.9742857142857144e-05, 'epoch': 4.21}\n",
      "{'loss': 0.0414, 'learning_rate': 5.938571428571429e-05, 'epoch': 4.24}\n",
      "{'loss': 0.0501, 'learning_rate': 5.902857142857143e-05, 'epoch': 4.27}\n",
      "{'loss': 0.0478, 'learning_rate': 5.867142857142858e-05, 'epoch': 4.3}\n",
      "{'loss': 0.0482, 'learning_rate': 5.8314285714285724e-05, 'epoch': 4.32}\n",
      "{'loss': 0.0463, 'learning_rate': 5.7957142857142864e-05, 'epoch': 4.35}\n",
      "{'loss': 0.0513, 'learning_rate': 5.76e-05, 'epoch': 4.38}\n",
      "{'loss': 0.0546, 'learning_rate': 5.7242857142857144e-05, 'epoch': 4.41}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5b1ec44b408a4de8aa8a3e9702dae453",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.28944167494773865, 'eval_wer': 21.885521885521886, 'eval_runtime': 344.5818, 'eval_samples_per_second': 2.206, 'eval_steps_per_second': 0.276, 'epoch': 4.41}\n",
      "{'loss': 0.0515, 'learning_rate': 5.6885714285714284e-05, 'epoch': 4.43}\n",
      "{'loss': 0.0394, 'learning_rate': 5.652857142857143e-05, 'epoch': 4.46}\n",
      "{'loss': 0.0562, 'learning_rate': 5.617142857142858e-05, 'epoch': 4.49}\n",
      "{'loss': 0.0532, 'learning_rate': 5.581428571428572e-05, 'epoch': 4.52}\n",
      "{'loss': 0.0525, 'learning_rate': 5.5457142857142864e-05, 'epoch': 4.54}\n",
      "{'loss': 0.0553, 'learning_rate': 5.5100000000000004e-05, 'epoch': 4.57}\n",
      "{'loss': 0.0464, 'learning_rate': 5.474285714285714e-05, 'epoch': 4.6}\n",
      "{'loss': 0.0425, 'learning_rate': 5.4385714285714284e-05, 'epoch': 4.63}\n",
      "{'loss': 0.0529, 'learning_rate': 5.402857142857143e-05, 'epoch': 4.65}\n",
      "{'loss': 0.0534, 'learning_rate': 5.367142857142857e-05, 'epoch': 4.68}\n",
      "{'loss': 0.0505, 'learning_rate': 5.331428571428572e-05, 'epoch': 4.71}\n",
      "{'loss': 0.0416, 'learning_rate': 5.295714285714286e-05, 'epoch': 4.74}\n",
      "{'loss': 0.0438, 'learning_rate': 5.2600000000000005e-05, 'epoch': 4.76}\n",
      "{'loss': 0.0568, 'learning_rate': 5.224285714285715e-05, 'epoch': 4.79}\n",
      "{'loss': 0.0519, 'learning_rate': 5.188571428571429e-05, 'epoch': 4.82}\n",
      "{'loss': 0.0415, 'learning_rate': 5.1528571428571425e-05, 'epoch': 4.85}\n",
      "{'loss': 0.0502, 'learning_rate': 5.117142857142857e-05, 'epoch': 4.87}\n",
      "{'loss': 0.0433, 'learning_rate': 5.081428571428571e-05, 'epoch': 4.9}\n",
      "{'loss': 0.0527, 'learning_rate': 5.045714285714286e-05, 'epoch': 4.93}\n",
      "{'loss': 0.0434, 'learning_rate': 5.0100000000000005e-05, 'epoch': 4.96}\n",
      "{'loss': 0.0485, 'learning_rate': 4.9742857142857145e-05, 'epoch': 4.98}\n",
      "{'loss': 0.0358, 'learning_rate': 4.938571428571429e-05, 'epoch': 5.01}\n",
      "{'loss': 0.0218, 'learning_rate': 4.902857142857143e-05, 'epoch': 5.04}\n",
      "{'loss': 0.0245, 'learning_rate': 4.867142857142857e-05, 'epoch': 5.07}\n",
      "{'loss': 0.0272, 'learning_rate': 4.831428571428572e-05, 'epoch': 5.09}\n",
      "{'loss': 0.0258, 'learning_rate': 4.795714285714286e-05, 'epoch': 5.12}\n",
      "{'loss': 0.0228, 'learning_rate': 4.76e-05, 'epoch': 5.15}\n",
      "{'loss': 0.0275, 'learning_rate': 4.7242857142857145e-05, 'epoch': 5.18}\n",
      "{'loss': 0.0269, 'learning_rate': 4.6885714285714285e-05, 'epoch': 5.2}\n",
      "{'loss': 0.0237, 'learning_rate': 4.652857142857143e-05, 'epoch': 5.23}\n",
      "{'loss': 0.0288, 'learning_rate': 4.617142857142857e-05, 'epoch': 5.26}\n",
      "{'loss': 0.0269, 'learning_rate': 4.581428571428572e-05, 'epoch': 5.29}\n",
      "{'loss': 0.0276, 'learning_rate': 4.545714285714286e-05, 'epoch': 5.31}\n",
      "{'loss': 0.0276, 'learning_rate': 4.5100000000000005e-05, 'epoch': 5.34}\n",
      "{'loss': 0.0242, 'learning_rate': 4.4742857142857145e-05, 'epoch': 5.37}\n",
      "{'loss': 0.0238, 'learning_rate': 4.4385714285714285e-05, 'epoch': 5.4}\n",
      "{'loss': 0.0302, 'learning_rate': 4.402857142857143e-05, 'epoch': 5.42}\n",
      "{'loss': 0.0253, 'learning_rate': 4.367142857142857e-05, 'epoch': 5.45}\n",
      "{'loss': 0.0256, 'learning_rate': 4.331428571428572e-05, 'epoch': 5.48}\n",
      "{'loss': 0.0256, 'learning_rate': 4.295714285714286e-05, 'epoch': 5.51}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2a3e7ace41cd44768ebc093aa571360e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.3023395836353302, 'eval_wer': 32.2973322973323, 'eval_runtime': 361.3589, 'eval_samples_per_second': 2.103, 'eval_steps_per_second': 0.263, 'epoch': 5.51}\n",
      "{'loss': 0.0215, 'learning_rate': 4.26e-05, 'epoch': 5.53}\n",
      "{'loss': 0.0272, 'learning_rate': 4.2242857142857145e-05, 'epoch': 5.56}\n",
      "{'loss': 0.0268, 'learning_rate': 4.188571428571429e-05, 'epoch': 5.59}\n",
      "{'loss': 0.028, 'learning_rate': 4.1528571428571425e-05, 'epoch': 5.62}\n",
      "{'loss': 0.0209, 'learning_rate': 4.117142857142857e-05, 'epoch': 5.64}\n",
      "{'loss': 0.0258, 'learning_rate': 4.081428571428572e-05, 'epoch': 5.67}\n",
      "{'loss': 0.0249, 'learning_rate': 4.045714285714286e-05, 'epoch': 5.7}\n",
      "{'loss': 0.0249, 'learning_rate': 4.0100000000000006e-05, 'epoch': 5.73}\n",
      "{'loss': 0.0209, 'learning_rate': 3.9742857142857146e-05, 'epoch': 5.75}\n",
      "{'loss': 0.02, 'learning_rate': 3.9385714285714286e-05, 'epoch': 5.78}\n",
      "{'loss': 0.0244, 'learning_rate': 3.902857142857143e-05, 'epoch': 5.81}\n",
      "{'loss': 0.025, 'learning_rate': 3.867142857142857e-05, 'epoch': 5.84}\n",
      "{'loss': 0.0282, 'learning_rate': 3.831428571428571e-05, 'epoch': 5.86}\n",
      "{'loss': 0.0271, 'learning_rate': 3.795714285714286e-05, 'epoch': 5.89}\n",
      "{'loss': 0.0233, 'learning_rate': 3.76e-05, 'epoch': 5.92}\n",
      "{'loss': 0.0219, 'learning_rate': 3.7242857142857146e-05, 'epoch': 5.95}\n",
      "{'loss': 0.0232, 'learning_rate': 3.688571428571429e-05, 'epoch': 5.97}\n",
      "{'loss': 0.019, 'learning_rate': 3.6528571428571426e-05, 'epoch': 6.0}\n",
      "{'loss': 0.0152, 'learning_rate': 3.617142857142857e-05, 'epoch': 6.03}\n",
      "{'loss': 0.0111, 'learning_rate': 3.581428571428572e-05, 'epoch': 6.06}\n",
      "{'loss': 0.0162, 'learning_rate': 3.545714285714286e-05, 'epoch': 6.08}\n",
      "{'loss': 0.0126, 'learning_rate': 3.51e-05, 'epoch': 6.11}\n",
      "{'loss': 0.012, 'learning_rate': 3.4742857142857146e-05, 'epoch': 6.14}\n",
      "{'loss': 0.0153, 'learning_rate': 3.4385714285714286e-05, 'epoch': 6.17}\n",
      "{'loss': 0.0133, 'learning_rate': 3.402857142857143e-05, 'epoch': 6.19}\n",
      "{'loss': 0.0112, 'learning_rate': 3.367142857142857e-05, 'epoch': 6.22}\n",
      "{'loss': 0.0187, 'learning_rate': 3.331428571428571e-05, 'epoch': 6.25}\n",
      "{'loss': 0.0134, 'learning_rate': 3.295714285714286e-05, 'epoch': 6.28}\n",
      "{'loss': 0.0112, 'learning_rate': 3.26e-05, 'epoch': 6.31}\n",
      "{'loss': 0.0096, 'learning_rate': 3.2242857142857146e-05, 'epoch': 6.33}\n",
      "{'loss': 0.0112, 'learning_rate': 3.1885714285714286e-05, 'epoch': 6.36}\n",
      "{'loss': 0.0146, 'learning_rate': 3.1528571428571426e-05, 'epoch': 6.39}\n",
      "{'loss': 0.0106, 'learning_rate': 3.117142857142857e-05, 'epoch': 6.42}\n",
      "{'loss': 0.01, 'learning_rate': 3.081428571428572e-05, 'epoch': 6.44}\n",
      "{'loss': 0.0117, 'learning_rate': 3.0457142857142856e-05, 'epoch': 6.47}\n",
      "{'loss': 0.0135, 'learning_rate': 3.01e-05, 'epoch': 6.5}\n",
      "{'loss': 0.0137, 'learning_rate': 2.9742857142857143e-05, 'epoch': 6.53}\n",
      "{'loss': 0.0089, 'learning_rate': 2.938571428571429e-05, 'epoch': 6.55}\n",
      "{'loss': 0.0096, 'learning_rate': 2.9028571428571427e-05, 'epoch': 6.58}\n",
      "{'loss': 0.0111, 'learning_rate': 2.867142857142857e-05, 'epoch': 6.61}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a911edc03d0b43edbd4c59958024dca9",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.3060542345046997, 'eval_wer': 31.015281015281015, 'eval_runtime': 366.4932, 'eval_samples_per_second': 2.074, 'eval_steps_per_second': 0.259, 'epoch': 6.61}\n",
      "{'loss': 0.0091, 'learning_rate': 2.8314285714285717e-05, 'epoch': 6.64}\n",
      "{'loss': 0.0075, 'learning_rate': 2.795714285714286e-05, 'epoch': 6.66}\n",
      "{'loss': 0.0096, 'learning_rate': 2.7600000000000003e-05, 'epoch': 6.69}\n",
      "{'loss': 0.0071, 'learning_rate': 2.7242857142857143e-05, 'epoch': 6.72}\n",
      "{'loss': 0.0089, 'learning_rate': 2.6885714285714287e-05, 'epoch': 6.75}\n",
      "{'loss': 0.0103, 'learning_rate': 2.652857142857143e-05, 'epoch': 6.77}\n",
      "{'loss': 0.0125, 'learning_rate': 2.6171428571428574e-05, 'epoch': 6.8}\n",
      "{'loss': 0.0082, 'learning_rate': 2.5814285714285713e-05, 'epoch': 6.83}\n",
      "{'loss': 0.0079, 'learning_rate': 2.5457142857142857e-05, 'epoch': 6.86}\n",
      "{'loss': 0.0108, 'learning_rate': 2.51e-05, 'epoch': 6.88}\n",
      "{'loss': 0.0084, 'learning_rate': 2.4742857142857147e-05, 'epoch': 6.91}\n",
      "{'loss': 0.0107, 'learning_rate': 2.4385714285714287e-05, 'epoch': 6.94}\n",
      "{'loss': 0.009, 'learning_rate': 2.402857142857143e-05, 'epoch': 6.97}\n",
      "{'loss': 0.0081, 'learning_rate': 2.3671428571428574e-05, 'epoch': 6.99}\n",
      "{'loss': 0.0077, 'learning_rate': 2.3314285714285717e-05, 'epoch': 7.02}\n",
      "{'loss': 0.0064, 'learning_rate': 2.2957142857142857e-05, 'epoch': 7.05}\n",
      "{'loss': 0.0079, 'learning_rate': 2.26e-05, 'epoch': 7.08}\n",
      "{'loss': 0.0063, 'learning_rate': 2.2242857142857144e-05, 'epoch': 7.1}\n",
      "{'loss': 0.0044, 'learning_rate': 2.1885714285714287e-05, 'epoch': 7.13}\n",
      "{'loss': 0.0041, 'learning_rate': 2.1528571428571427e-05, 'epoch': 7.16}\n",
      "{'loss': 0.0048, 'learning_rate': 2.1171428571428574e-05, 'epoch': 7.19}\n",
      "{'loss': 0.0041, 'learning_rate': 2.0814285714285714e-05, 'epoch': 7.21}\n",
      "{'loss': 0.0031, 'learning_rate': 2.0457142857142857e-05, 'epoch': 7.24}\n",
      "{'loss': 0.0026, 'learning_rate': 2.01e-05, 'epoch': 7.27}\n",
      "{'loss': 0.0031, 'learning_rate': 1.9742857142857144e-05, 'epoch': 7.3}\n",
      "{'loss': 0.0029, 'learning_rate': 1.9385714285714287e-05, 'epoch': 7.32}\n",
      "{'loss': 0.0045, 'learning_rate': 1.9028571428571427e-05, 'epoch': 7.35}\n",
      "{'loss': 0.0024, 'learning_rate': 1.8671428571428574e-05, 'epoch': 7.38}\n",
      "{'loss': 0.002, 'learning_rate': 1.8314285714285714e-05, 'epoch': 7.41}\n",
      "{'loss': 0.0038, 'learning_rate': 1.7957142857142858e-05, 'epoch': 7.43}\n",
      "{'loss': 0.0035, 'learning_rate': 1.76e-05, 'epoch': 7.46}\n",
      "{'loss': 0.0058, 'learning_rate': 1.7242857142857144e-05, 'epoch': 7.49}\n",
      "{'loss': 0.0034, 'learning_rate': 1.6885714285714284e-05, 'epoch': 7.52}\n",
      "{'loss': 0.0036, 'learning_rate': 1.652857142857143e-05, 'epoch': 7.54}\n",
      "{'loss': 0.0031, 'learning_rate': 1.6171428571428574e-05, 'epoch': 7.57}\n",
      "{'loss': 0.0041, 'learning_rate': 1.5814285714285714e-05, 'epoch': 7.6}\n",
      "{'loss': 0.0021, 'learning_rate': 1.5457142857142858e-05, 'epoch': 7.63}\n",
      "{'loss': 0.0032, 'learning_rate': 1.51e-05, 'epoch': 7.65}\n",
      "{'loss': 0.0039, 'learning_rate': 1.4742857142857144e-05, 'epoch': 7.68}\n",
      "{'loss': 0.0028, 'learning_rate': 1.4385714285714286e-05, 'epoch': 7.71}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3ff56df660d1427daf8f920cd57d67fc",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.3143082559108734, 'eval_wer': 27.169127169127172, 'eval_runtime': 357.6908, 'eval_samples_per_second': 2.125, 'eval_steps_per_second': 0.266, 'epoch': 7.71}\n",
      "{'loss': 0.0032, 'learning_rate': 1.402857142857143e-05, 'epoch': 7.74}\n",
      "{'loss': 0.0028, 'learning_rate': 1.3671428571428571e-05, 'epoch': 7.76}\n",
      "{'loss': 0.0033, 'learning_rate': 1.3314285714285715e-05, 'epoch': 7.79}\n",
      "{'loss': 0.0052, 'learning_rate': 1.2957142857142856e-05, 'epoch': 7.82}\n",
      "{'loss': 0.0023, 'learning_rate': 1.2600000000000001e-05, 'epoch': 7.85}\n",
      "{'loss': 0.0034, 'learning_rate': 1.2242857142857143e-05, 'epoch': 7.87}\n",
      "{'loss': 0.0026, 'learning_rate': 1.1885714285714286e-05, 'epoch': 7.9}\n",
      "{'loss': 0.0022, 'learning_rate': 1.1528571428571428e-05, 'epoch': 7.93}\n",
      "{'loss': 0.0037, 'learning_rate': 1.1171428571428571e-05, 'epoch': 7.96}\n",
      "{'loss': 0.003, 'learning_rate': 1.0814285714285715e-05, 'epoch': 7.98}\n",
      "{'loss': 0.0009, 'learning_rate': 1.0457142857142856e-05, 'epoch': 8.01}\n",
      "{'loss': 0.0006, 'learning_rate': 1.0100000000000002e-05, 'epoch': 8.04}\n",
      "{'loss': 0.0022, 'learning_rate': 9.742857142857143e-06, 'epoch': 8.07}\n",
      "{'loss': 0.0005, 'learning_rate': 9.385714285714287e-06, 'epoch': 8.09}\n",
      "{'loss': 0.0007, 'learning_rate': 9.02857142857143e-06, 'epoch': 8.12}\n",
      "{'loss': 0.0006, 'learning_rate': 8.671428571428572e-06, 'epoch': 8.15}\n",
      "{'loss': 0.0006, 'learning_rate': 8.314285714285715e-06, 'epoch': 8.18}\n",
      "{'loss': 0.0005, 'learning_rate': 7.957142857142857e-06, 'epoch': 8.2}\n",
      "{'loss': 0.0024, 'learning_rate': 7.6e-06, 'epoch': 8.23}\n",
      "{'loss': 0.0009, 'learning_rate': 7.242857142857143e-06, 'epoch': 8.26}\n",
      "{'loss': 0.0004, 'learning_rate': 6.885714285714286e-06, 'epoch': 8.29}\n",
      "{'loss': 0.0006, 'learning_rate': 6.5285714285714285e-06, 'epoch': 8.31}\n",
      "{'loss': 0.0031, 'learning_rate': 6.171428571428572e-06, 'epoch': 8.34}\n",
      "{'loss': 0.001, 'learning_rate': 5.814285714285714e-06, 'epoch': 8.37}\n",
      "{'loss': 0.0013, 'learning_rate': 5.457142857142857e-06, 'epoch': 8.4}\n",
      "{'loss': 0.0011, 'learning_rate': 5.1e-06, 'epoch': 8.43}\n",
      "{'loss': 0.0004, 'learning_rate': 4.742857142857144e-06, 'epoch': 8.45}\n",
      "{'loss': 0.0012, 'learning_rate': 4.385714285714286e-06, 'epoch': 8.48}\n",
      "{'loss': 0.0013, 'learning_rate': 4.028571428571429e-06, 'epoch': 8.51}\n",
      "{'loss': 0.0008, 'learning_rate': 3.6714285714285717e-06, 'epoch': 8.54}\n",
      "{'loss': 0.0009, 'learning_rate': 3.314285714285714e-06, 'epoch': 8.56}\n",
      "{'loss': 0.0019, 'learning_rate': 2.957142857142857e-06, 'epoch': 8.59}\n",
      "{'loss': 0.0006, 'learning_rate': 2.6e-06, 'epoch': 8.62}\n",
      "{'loss': 0.0004, 'learning_rate': 2.242857142857143e-06, 'epoch': 8.65}\n",
      "{'loss': 0.0005, 'learning_rate': 1.8857142857142858e-06, 'epoch': 8.67}\n",
      "{'loss': 0.0005, 'learning_rate': 1.5285714285714287e-06, 'epoch': 8.7}\n",
      "{'loss': 0.001, 'learning_rate': 1.1714285714285715e-06, 'epoch': 8.73}\n",
      "{'loss': 0.0012, 'learning_rate': 8.142857142857143e-07, 'epoch': 8.76}\n",
      "{'loss': 0.001, 'learning_rate': 4.571428571428572e-07, 'epoch': 8.78}\n",
      "{'loss': 0.0014, 'learning_rate': 1.0000000000000001e-07, 'epoch': 8.81}\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5120f3a800554c1689dfb561c55440fb",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/95 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'eval_loss': 0.318661630153656, 'eval_wer': 27.36337736337736, 'eval_runtime': 356.3989, 'eval_samples_per_second': 2.132, 'eval_steps_per_second': 0.267, 'epoch': 8.81}\n",
      "{'train_runtime': 48216.7702, 'train_samples_per_second': 2.655, 'train_steps_per_second': 0.166, 'train_loss': 0.13303363310021815, 'epoch': 8.81}\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "TrainOutput(global_step=8000, training_loss=0.13303363310021815, metrics={'train_runtime': 48216.7702, 'train_samples_per_second': 2.655, 'train_steps_per_second': 0.166, 'train_loss': 0.13303363310021815, 'epoch': 8.81})"
      ]
     },
     "execution_count": 111,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "ename": "OSError",
     "evalue": "It looks like the config file at './whisper-base-vi/pytorch_model.bin' is not a valid JSON file.",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mUnicodeDecodeError\u001b[0m                        Traceback (most recent call last)",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/configuration_utils.py:702\u001b[0m, in \u001b[0;36mPretrainedConfig._get_config_dict\u001b[0;34m(cls, pretrained_model_name_or_path, **kwargs)\u001b[0m\n\u001b[1;32m    700\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[1;32m    701\u001b[0m     \u001b[39m# Load config dict\u001b[39;00m\n\u001b[0;32m--> 702\u001b[0m     config_dict \u001b[39m=\u001b[39m \u001b[39mcls\u001b[39;49m\u001b[39m.\u001b[39;49m_dict_from_json_file(resolved_config_file)\n\u001b[1;32m    703\u001b[0m     config_dict[\u001b[39m\"\u001b[39m\u001b[39m_commit_hash\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m commit_hash\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/configuration_utils.py:793\u001b[0m, in \u001b[0;36mPretrainedConfig._dict_from_json_file\u001b[0;34m(cls, json_file)\u001b[0m\n\u001b[1;32m    792\u001b[0m \u001b[39mwith\u001b[39;00m \u001b[39mopen\u001b[39m(json_file, \u001b[39m\"\u001b[39m\u001b[39mr\u001b[39m\u001b[39m\"\u001b[39m, encoding\u001b[39m=\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mutf-8\u001b[39m\u001b[39m\"\u001b[39m) \u001b[39mas\u001b[39;00m reader:\n\u001b[0;32m--> 793\u001b[0m     text \u001b[39m=\u001b[39m reader\u001b[39m.\u001b[39;49mread()\n\u001b[1;32m    794\u001b[0m \u001b[39mreturn\u001b[39;00m json\u001b[39m.\u001b[39mloads(text)\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/codecs.py:322\u001b[0m, in \u001b[0;36mBufferedIncrementalDecoder.decode\u001b[0;34m(self, input, final)\u001b[0m\n\u001b[1;32m    321\u001b[0m data \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mbuffer \u001b[39m+\u001b[39m \u001b[39minput\u001b[39m\n\u001b[0;32m--> 322\u001b[0m (result, consumed) \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_buffer_decode(data, \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49merrors, final)\n\u001b[1;32m    323\u001b[0m \u001b[39m# keep undecoded input until the next call\u001b[39;00m\n",
      "\u001b[0;31mUnicodeDecodeError\u001b[0m: 'utf-8' codec can't decode byte 0x80 in position 64: invalid start byte",
      "\nDuring handling of the above exception, another exception occurred:\n",
      "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[34], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m pt_model \u001b[39m=\u001b[39m WhisperForConditionalGeneration\u001b[39m.\u001b[39;49mfrom_pretrained(\u001b[39m\"\u001b[39;49m\u001b[39m./whisper-base-vi/pytorch_model.bin\u001b[39;49m\u001b[39m\"\u001b[39;49m, from_tf\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m)\n\u001b[1;32m      2\u001b[0m pt_model\u001b[39m.\u001b[39msave_pretrained(\u001b[39m\"\u001b[39m\u001b[39m./whisper-base-vi/vi_whisper.pt\u001b[39m\u001b[39m\"\u001b[39m)\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/modeling_utils.py:2325\u001b[0m, in \u001b[0;36mPreTrainedModel.from_pretrained\u001b[0;34m(cls, pretrained_model_name_or_path, config, cache_dir, ignore_mismatched_sizes, force_download, local_files_only, token, revision, use_safetensors, *model_args, **kwargs)\u001b[0m\n\u001b[1;32m   2323\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39misinstance\u001b[39m(config, PretrainedConfig):\n\u001b[1;32m   2324\u001b[0m     config_path \u001b[39m=\u001b[39m config \u001b[39mif\u001b[39;00m config \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m \u001b[39melse\u001b[39;00m pretrained_model_name_or_path\n\u001b[0;32m-> 2325\u001b[0m     config, model_kwargs \u001b[39m=\u001b[39m \u001b[39mcls\u001b[39;49m\u001b[39m.\u001b[39;49mconfig_class\u001b[39m.\u001b[39;49mfrom_pretrained(\n\u001b[1;32m   2326\u001b[0m         config_path,\n\u001b[1;32m   2327\u001b[0m         cache_dir\u001b[39m=\u001b[39;49mcache_dir,\n\u001b[1;32m   2328\u001b[0m         return_unused_kwargs\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m,\n\u001b[1;32m   2329\u001b[0m         force_download\u001b[39m=\u001b[39;49mforce_download,\n\u001b[1;32m   2330\u001b[0m         resume_download\u001b[39m=\u001b[39;49mresume_download,\n\u001b[1;32m   2331\u001b[0m         proxies\u001b[39m=\u001b[39;49mproxies,\n\u001b[1;32m   2332\u001b[0m         local_files_only\u001b[39m=\u001b[39;49mlocal_files_only,\n\u001b[1;32m   2333\u001b[0m         token\u001b[39m=\u001b[39;49mtoken,\n\u001b[1;32m   2334\u001b[0m         revision\u001b[39m=\u001b[39;49mrevision,\n\u001b[1;32m   2335\u001b[0m         subfolder\u001b[39m=\u001b[39;49msubfolder,\n\u001b[1;32m   2336\u001b[0m         _from_auto\u001b[39m=\u001b[39;49mfrom_auto_class,\n\u001b[1;32m   2337\u001b[0m         _from_pipeline\u001b[39m=\u001b[39;49mfrom_pipeline,\n\u001b[1;32m   2338\u001b[0m         \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs,\n\u001b[1;32m   2339\u001b[0m     )\n\u001b[1;32m   2340\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[1;32m   2341\u001b[0m     model_kwargs \u001b[39m=\u001b[39m kwargs\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/configuration_utils.py:590\u001b[0m, in \u001b[0;36mPretrainedConfig.from_pretrained\u001b[0;34m(cls, pretrained_model_name_or_path, cache_dir, force_download, local_files_only, token, revision, **kwargs)\u001b[0m\n\u001b[1;32m    586\u001b[0m kwargs[\u001b[39m\"\u001b[39m\u001b[39mrevision\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m revision\n\u001b[1;32m    588\u001b[0m \u001b[39mcls\u001b[39m\u001b[39m.\u001b[39m_set_token_in_kwargs(kwargs, token)\n\u001b[0;32m--> 590\u001b[0m config_dict, kwargs \u001b[39m=\u001b[39m \u001b[39mcls\u001b[39;49m\u001b[39m.\u001b[39;49mget_config_dict(pretrained_model_name_or_path, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m    591\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39m\"\u001b[39m\u001b[39mmodel_type\u001b[39m\u001b[39m\"\u001b[39m \u001b[39min\u001b[39;00m config_dict \u001b[39mand\u001b[39;00m \u001b[39mhasattr\u001b[39m(\u001b[39mcls\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39mmodel_type\u001b[39m\u001b[39m\"\u001b[39m) \u001b[39mand\u001b[39;00m config_dict[\u001b[39m\"\u001b[39m\u001b[39mmodel_type\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m!=\u001b[39m \u001b[39mcls\u001b[39m\u001b[39m.\u001b[39mmodel_type:\n\u001b[1;32m    592\u001b[0m     logger\u001b[39m.\u001b[39mwarning(\n\u001b[1;32m    593\u001b[0m         \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mYou are using a model of type \u001b[39m\u001b[39m{\u001b[39;00mconfig_dict[\u001b[39m'\u001b[39m\u001b[39mmodel_type\u001b[39m\u001b[39m'\u001b[39m]\u001b[39m}\u001b[39;00m\u001b[39m to instantiate a model of type \u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    594\u001b[0m         \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39m{\u001b[39;00m\u001b[39mcls\u001b[39m\u001b[39m.\u001b[39mmodel_type\u001b[39m}\u001b[39;00m\u001b[39m. This is not supported for all configurations of models and can yield errors.\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    595\u001b[0m     )\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/configuration_utils.py:617\u001b[0m, in \u001b[0;36mPretrainedConfig.get_config_dict\u001b[0;34m(cls, pretrained_model_name_or_path, **kwargs)\u001b[0m\n\u001b[1;32m    615\u001b[0m original_kwargs \u001b[39m=\u001b[39m copy\u001b[39m.\u001b[39mdeepcopy(kwargs)\n\u001b[1;32m    616\u001b[0m \u001b[39m# Get config dict associated with the base config file\u001b[39;00m\n\u001b[0;32m--> 617\u001b[0m config_dict, kwargs \u001b[39m=\u001b[39m \u001b[39mcls\u001b[39;49m\u001b[39m.\u001b[39;49m_get_config_dict(pretrained_model_name_or_path, \u001b[39m*\u001b[39;49m\u001b[39m*\u001b[39;49mkwargs)\n\u001b[1;32m    618\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39m\"\u001b[39m\u001b[39m_commit_hash\u001b[39m\u001b[39m\"\u001b[39m \u001b[39min\u001b[39;00m config_dict:\n\u001b[1;32m    619\u001b[0m     original_kwargs[\u001b[39m\"\u001b[39m\u001b[39m_commit_hash\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m config_dict[\u001b[39m\"\u001b[39m\u001b[39m_commit_hash\u001b[39m\u001b[39m\"\u001b[39m]\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/configuration_utils.py:705\u001b[0m, in \u001b[0;36mPretrainedConfig._get_config_dict\u001b[0;34m(cls, pretrained_model_name_or_path, **kwargs)\u001b[0m\n\u001b[1;32m    703\u001b[0m     config_dict[\u001b[39m\"\u001b[39m\u001b[39m_commit_hash\u001b[39m\u001b[39m\"\u001b[39m] \u001b[39m=\u001b[39m commit_hash\n\u001b[1;32m    704\u001b[0m \u001b[39mexcept\u001b[39;00m (json\u001b[39m.\u001b[39mJSONDecodeError, \u001b[39mUnicodeDecodeError\u001b[39;00m):\n\u001b[0;32m--> 705\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mEnvironmentError\u001b[39;00m(\n\u001b[1;32m    706\u001b[0m         \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mIt looks like the config file at \u001b[39m\u001b[39m'\u001b[39m\u001b[39m{\u001b[39;00mresolved_config_file\u001b[39m}\u001b[39;00m\u001b[39m'\u001b[39m\u001b[39m is not a valid JSON file.\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    707\u001b[0m     )\n\u001b[1;32m    709\u001b[0m \u001b[39mif\u001b[39;00m is_local:\n\u001b[1;32m    710\u001b[0m     logger\u001b[39m.\u001b[39minfo(\u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mloading configuration file \u001b[39m\u001b[39m{\u001b[39;00mresolved_config_file\u001b[39m}\u001b[39;00m\u001b[39m\"\u001b[39m)\n",
      "\u001b[0;31mOSError\u001b[0m: It looks like the config file at './whisper-base-vi/pytorch_model.bin' is not a valid JSON file."
     ]
    }
   ],
   "source": [
    "pt_model = WhisperForConditionalGeneration.from_pretrained(\"./whisper-base-vi/pytorch_model.bin\", from_tf=True)\n",
    "pt_model.save_pretrained(\"./whisper-base-vi/vi_whisper.pt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "kwargs = {\n",
    "    \"dataset_tags\": \"vivos-commonvoice\",\n",
    "    \"dataset\": \"Vivos\",  \n",
    "    \"language\": \"vi\",\n",
    "    \"model_name\": \"Whisper Small Vi - Duy Ta\", \n",
    "    \"finetuned_from\": \"openai/whisper-small\",\n",
    "    \"tasks\": \"automatic-speech-recognition\",\n",
    "    \"config\" : None\n",
    "}\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Several commits (2) will be pushed upstream.\n",
      "The progress bars may be unreliable.\n",
      "error: The destination you provided is not a full refname (i.e.,\n",
      "starting with \"refs/\"). We tried to guess what you meant by:\n",
      "\n",
      "- Looking for a ref that matches 'HEAD' on the remote side.\n",
      "- Checking if the <src> being pushed ('HEAD')\n",
      "  is a ref in \"refs/{heads,tags}/\". If so we add a corresponding\n",
      "  refs/{heads,tags}/ prefix on the remote side.\n",
      "\n",
      "Neither worked, so we gave up. You must fully qualify the ref.\n",
      "hint: The <src> part of the refspec is a commit object.\n",
      "hint: Did you mean to create a new branch by pushing to\n",
      "hint: 'HEAD:refs/heads/HEAD'?\n",
      "error: failed to push some refs to 'https://huggingface.co/DuyTa/vi_whisper-small'\n",
      "\n"
     ]
    },
    {
     "ename": "OSError",
     "evalue": "error: The destination you provided is not a full refname (i.e.,\nstarting with \"refs/\"). We tried to guess what you meant by:\n\n- Looking for a ref that matches 'HEAD' on the remote side.\n- Checking if the <src> being pushed ('HEAD')\n  is a ref in \"refs/{heads,tags}/\". If so we add a corresponding\n  refs/{heads,tags}/ prefix on the remote side.\n\nNeither worked, so we gave up. You must fully qualify the ref.\nhint: The <src> part of the refspec is a commit object.\nhint: Did you mean to create a new branch by pushing to\nhint: 'HEAD:refs/heads/HEAD'?\nerror: failed to push some refs to 'https://huggingface.co/DuyTa/vi_whisper-small'\n",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mCalledProcessError\u001b[0m                        Traceback (most recent call last)",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/repository.py:1099\u001b[0m, in \u001b[0;36mRepository.git_push\u001b[0;34m(self, upstream, blocking, auto_lfs_prune)\u001b[0m\n\u001b[1;32m   1098\u001b[0m             \u001b[39mif\u001b[39;00m return_code:\n\u001b[0;32m-> 1099\u001b[0m                 \u001b[39mraise\u001b[39;00m subprocess\u001b[39m.\u001b[39mCalledProcessError(return_code, process\u001b[39m.\u001b[39margs, output\u001b[39m=\u001b[39mstdout, stderr\u001b[39m=\u001b[39mstderr)\n\u001b[1;32m   1101\u001b[0m \u001b[39mexcept\u001b[39;00m subprocess\u001b[39m.\u001b[39mCalledProcessError \u001b[39mas\u001b[39;00m exc:\n",
      "\u001b[0;31mCalledProcessError\u001b[0m: Command '['git', 'push', '--set-upstream', 'origin', 'HEAD']' returned non-zero exit status 1.",
      "\nDuring handling of the above exception, another exception occurred:\n",
      "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[131], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m trainer\u001b[39m.\u001b[39;49mpush_to_hub(commit_message\u001b[39m=\u001b[39;49m\u001b[39m\"\u001b[39;49m\u001b[39mchange\u001b[39;49m\u001b[39m\"\u001b[39;49m)\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/trainer.py:3609\u001b[0m, in \u001b[0;36mTrainer.push_to_hub\u001b[0;34m(self, commit_message, blocking, **kwargs)\u001b[0m\n\u001b[1;32m   3606\u001b[0m     \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mpush_in_progress\u001b[39m.\u001b[39m_process\u001b[39m.\u001b[39mkill()\n\u001b[1;32m   3607\u001b[0m     \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mpush_in_progress \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[0;32m-> 3609\u001b[0m git_head_commit_url \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mrepo\u001b[39m.\u001b[39;49mpush_to_hub(\n\u001b[1;32m   3610\u001b[0m     commit_message\u001b[39m=\u001b[39;49mcommit_message, blocking\u001b[39m=\u001b[39;49mblocking, auto_lfs_prune\u001b[39m=\u001b[39;49m\u001b[39mTrue\u001b[39;49;00m\n\u001b[1;32m   3611\u001b[0m )\n\u001b[1;32m   3612\u001b[0m \u001b[39m# push separately the model card to be independant from the rest of the model\u001b[39;00m\n\u001b[1;32m   3613\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39margs\u001b[39m.\u001b[39mshould_save:\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/repository.py:1307\u001b[0m, in \u001b[0;36mRepository.push_to_hub\u001b[0;34m(self, commit_message, blocking, clean_ok, auto_lfs_prune)\u001b[0m\n\u001b[1;32m   1305\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mgit_add(auto_lfs_track\u001b[39m=\u001b[39m\u001b[39mTrue\u001b[39;00m)\n\u001b[1;32m   1306\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39mgit_commit(commit_message)\n\u001b[0;32m-> 1307\u001b[0m \u001b[39mreturn\u001b[39;00m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mgit_push(\n\u001b[1;32m   1308\u001b[0m     upstream\u001b[39m=\u001b[39;49m\u001b[39mf\u001b[39;49m\u001b[39m\"\u001b[39;49m\u001b[39morigin \u001b[39;49m\u001b[39m{\u001b[39;49;00m\u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49mcurrent_branch\u001b[39m}\u001b[39;49;00m\u001b[39m\"\u001b[39;49m,\n\u001b[1;32m   1309\u001b[0m     blocking\u001b[39m=\u001b[39;49mblocking,\n\u001b[1;32m   1310\u001b[0m     auto_lfs_prune\u001b[39m=\u001b[39;49mauto_lfs_prune,\n\u001b[1;32m   1311\u001b[0m )\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/repository.py:1102\u001b[0m, in \u001b[0;36mRepository.git_push\u001b[0;34m(self, upstream, blocking, auto_lfs_prune)\u001b[0m\n\u001b[1;32m   1099\u001b[0m                 \u001b[39mraise\u001b[39;00m subprocess\u001b[39m.\u001b[39mCalledProcessError(return_code, process\u001b[39m.\u001b[39margs, output\u001b[39m=\u001b[39mstdout, stderr\u001b[39m=\u001b[39mstderr)\n\u001b[1;32m   1101\u001b[0m \u001b[39mexcept\u001b[39;00m subprocess\u001b[39m.\u001b[39mCalledProcessError \u001b[39mas\u001b[39;00m exc:\n\u001b[0;32m-> 1102\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mEnvironmentError\u001b[39;00m(exc\u001b[39m.\u001b[39mstderr)\n\u001b[1;32m   1104\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m blocking:\n\u001b[1;32m   1106\u001b[0m     \u001b[39mdef\u001b[39;00m \u001b[39mstatus_method\u001b[39m():\n",
      "\u001b[0;31mOSError\u001b[0m: error: The destination you provided is not a full refname (i.e.,\nstarting with \"refs/\"). We tried to guess what you meant by:\n\n- Looking for a ref that matches 'HEAD' on the remote side.\n- Checking if the <src> being pushed ('HEAD')\n  is a ref in \"refs/{heads,tags}/\". If so we add a corresponding\n  refs/{heads,tags}/ prefix on the remote side.\n\nNeither worked, so we gave up. You must fully qualify the ref.\nhint: The <src> part of the refspec is a commit object.\nhint: Did you mean to create a new branch by pushing to\nhint: 'HEAD:refs/heads/HEAD'?\nerror: failed to push some refs to 'https://huggingface.co/DuyTa/vi_whisper-small'\n"
     ]
    }
   ],
   "source": [
    "trainer.push_to_hub(commit_message=\"change\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "tags": [
     "parameters"
    ]
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b6e666bab7b2450abf3e2adf07679122",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading (…)lve/main/config.json:   0%|          | 0.00/1.31k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b212026dca9241cf994f9710f0b93c22",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading (…)okenizer_config.json:   0%|          | 0.00/838 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from transformers import WhisperForConditionalGeneration, WhisperProcessor\n",
    "\n",
    "model = WhisperForConditionalGeneration.from_pretrained(\"DuyTa/vi_whisper\")\n",
    "processor = WhisperProcessor.from_pretrained(\"DuyTa/vi_whisper\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "ename": "RuntimeError",
     "evalue": "Instantiating a pipeline without a task set raised an error: Repo id must use alphanumeric chars or '-', '_', '.', '--' and '..' are forbidden, '-' and '.' cannot start or end the name, max length is 96: './vi_whisper-small'.",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mHFValidationError\u001b[0m                         Traceback (most recent call last)",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/pipelines/__init__.py:432\u001b[0m, in \u001b[0;36mget_task\u001b[0;34m(model, use_auth_token)\u001b[0m\n\u001b[1;32m    431\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m--> 432\u001b[0m     info \u001b[39m=\u001b[39m model_info(model, token\u001b[39m=\u001b[39;49muse_auth_token)\n\u001b[1;32m    433\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e:\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/utils/_validators.py:110\u001b[0m, in \u001b[0;36mvalidate_hf_hub_args.<locals>._inner_fn\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    109\u001b[0m \u001b[39mif\u001b[39;00m arg_name \u001b[39min\u001b[39;00m [\u001b[39m\"\u001b[39m\u001b[39mrepo_id\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39mfrom_id\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m\"\u001b[39m\u001b[39mto_id\u001b[39m\u001b[39m\"\u001b[39m]:\n\u001b[0;32m--> 110\u001b[0m     validate_repo_id(arg_value)\n\u001b[1;32m    112\u001b[0m \u001b[39melif\u001b[39;00m arg_name \u001b[39m==\u001b[39m \u001b[39m\"\u001b[39m\u001b[39mtoken\u001b[39m\u001b[39m\"\u001b[39m \u001b[39mand\u001b[39;00m arg_value \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/huggingface_hub/utils/_validators.py:164\u001b[0m, in \u001b[0;36mvalidate_repo_id\u001b[0;34m(repo_id)\u001b[0m\n\u001b[1;32m    163\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m REPO_ID_REGEX\u001b[39m.\u001b[39mmatch(repo_id):\n\u001b[0;32m--> 164\u001b[0m     \u001b[39mraise\u001b[39;00m HFValidationError(\n\u001b[1;32m    165\u001b[0m         \u001b[39m\"\u001b[39m\u001b[39mRepo id must use alphanumeric chars or \u001b[39m\u001b[39m'\u001b[39m\u001b[39m-\u001b[39m\u001b[39m'\u001b[39m\u001b[39m, \u001b[39m\u001b[39m'\u001b[39m\u001b[39m_\u001b[39m\u001b[39m'\u001b[39m\u001b[39m, \u001b[39m\u001b[39m'\u001b[39m\u001b[39m.\u001b[39m\u001b[39m'\u001b[39m\u001b[39m, \u001b[39m\u001b[39m'\u001b[39m\u001b[39m--\u001b[39m\u001b[39m'\u001b[39m\u001b[39m and \u001b[39m\u001b[39m'\u001b[39m\u001b[39m..\u001b[39m\u001b[39m'\u001b[39m\u001b[39m are\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    166\u001b[0m         \u001b[39m\"\u001b[39m\u001b[39m forbidden, \u001b[39m\u001b[39m'\u001b[39m\u001b[39m-\u001b[39m\u001b[39m'\u001b[39m\u001b[39m and \u001b[39m\u001b[39m'\u001b[39m\u001b[39m.\u001b[39m\u001b[39m'\u001b[39m\u001b[39m cannot start or end the name, max length is 96:\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    167\u001b[0m         \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39m \u001b[39m\u001b[39m'\u001b[39m\u001b[39m{\u001b[39;00mrepo_id\u001b[39m}\u001b[39;00m\u001b[39m'\u001b[39m\u001b[39m.\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    168\u001b[0m     )\n\u001b[1;32m    170\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39m\"\u001b[39m\u001b[39m--\u001b[39m\u001b[39m\"\u001b[39m \u001b[39min\u001b[39;00m repo_id \u001b[39mor\u001b[39;00m \u001b[39m\"\u001b[39m\u001b[39m..\u001b[39m\u001b[39m\"\u001b[39m \u001b[39min\u001b[39;00m repo_id:\n",
      "\u001b[0;31mHFValidationError\u001b[0m: Repo id must use alphanumeric chars or '-', '_', '.', '--' and '..' are forbidden, '-' and '.' cannot start or end the name, max length is 96: './vi_whisper-small'.",
      "\nDuring handling of the above exception, another exception occurred:\n",
      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[36], line 4\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[39mfrom\u001b[39;00m \u001b[39mtransformers\u001b[39;00m \u001b[39mimport\u001b[39;00m pipeline\n\u001b[1;32m      2\u001b[0m \u001b[39mimport\u001b[39;00m \u001b[39mgradio\u001b[39;00m \u001b[39mas\u001b[39;00m \u001b[39mgr\u001b[39;00m\n\u001b[0;32m----> 4\u001b[0m pipe \u001b[39m=\u001b[39m pipeline(model\u001b[39m=\u001b[39;49m\u001b[39m\"\u001b[39;49m\u001b[39m./vi_whisper-small\u001b[39;49m\u001b[39m\"\u001b[39;49m)  \n\u001b[1;32m      6\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mtranscribe\u001b[39m(audio):\n\u001b[1;32m      7\u001b[0m     text \u001b[39m=\u001b[39m pipe(audio)[\u001b[39m\"\u001b[39m\u001b[39mtext\u001b[39m\u001b[39m\"\u001b[39m]\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/pipelines/__init__.py:726\u001b[0m, in \u001b[0;36mpipeline\u001b[0;34m(task, model, config, tokenizer, feature_extractor, image_processor, framework, revision, use_fast, use_auth_token, device, device_map, torch_dtype, trust_remote_code, model_kwargs, pipeline_class, **kwargs)\u001b[0m\n\u001b[1;32m    721\u001b[0m     \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39misinstance\u001b[39m(model, \u001b[39mstr\u001b[39m):\n\u001b[1;32m    722\u001b[0m         \u001b[39mraise\u001b[39;00m \u001b[39mRuntimeError\u001b[39;00m(\n\u001b[1;32m    723\u001b[0m             \u001b[39m\"\u001b[39m\u001b[39mInferring the task automatically requires to check the hub with a model_id defined as a `str`.\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    724\u001b[0m             \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39m{\u001b[39;00mmodel\u001b[39m}\u001b[39;00m\u001b[39m is not a valid model_id.\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    725\u001b[0m         )\n\u001b[0;32m--> 726\u001b[0m     task \u001b[39m=\u001b[39m get_task(model, use_auth_token)\n\u001b[1;32m    728\u001b[0m \u001b[39m# Retrieve the task\u001b[39;00m\n\u001b[1;32m    729\u001b[0m \u001b[39mif\u001b[39;00m task \u001b[39min\u001b[39;00m custom_tasks:\n",
      "File \u001b[0;32m~/miniconda3/envs/DUY/lib/python3.9/site-packages/transformers/pipelines/__init__.py:434\u001b[0m, in \u001b[0;36mget_task\u001b[0;34m(model, use_auth_token)\u001b[0m\n\u001b[1;32m    432\u001b[0m     info \u001b[39m=\u001b[39m model_info(model, token\u001b[39m=\u001b[39muse_auth_token)\n\u001b[1;32m    433\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mException\u001b[39;00m \u001b[39mas\u001b[39;00m e:\n\u001b[0;32m--> 434\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mRuntimeError\u001b[39;00m(\u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mInstantiating a pipeline without a task set raised an error: \u001b[39m\u001b[39m{\u001b[39;00me\u001b[39m}\u001b[39;00m\u001b[39m\"\u001b[39m)\n\u001b[1;32m    435\u001b[0m \u001b[39mif\u001b[39;00m \u001b[39mnot\u001b[39;00m info\u001b[39m.\u001b[39mpipeline_tag:\n\u001b[1;32m    436\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mRuntimeError\u001b[39;00m(\n\u001b[1;32m    437\u001b[0m         \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mThe model \u001b[39m\u001b[39m{\u001b[39;00mmodel\u001b[39m}\u001b[39;00m\u001b[39m does not seem to have a correct `pipeline_tag` set to infer the task automatically\u001b[39m\u001b[39m\"\u001b[39m\n\u001b[1;32m    438\u001b[0m     )\n",
      "\u001b[0;31mRuntimeError\u001b[0m: Instantiating a pipeline without a task set raised an error: Repo id must use alphanumeric chars or '-', '_', '.', '--' and '..' are forbidden, '-' and '.' cannot start or end the name, max length is 96: './vi_whisper-small'."
     ]
    }
   ],
   "source": [
    "from transformers import pipeline\n",
    "import gradio as gr\n",
    "\n",
    "pipe = pipeline(model=\"./vi_whisper-small\")  \n",
    "\n",
    "def transcribe(audio):\n",
    "    text = pipe(audio)[\"text\"]\n",
    "    return text\n",
    "\n",
    "iface = gr.Interface(\n",
    "    fn=transcribe,\n",
    "    inputs=gr.Audio(source=\"upload\", type=\"filepath\"),\n",
    "    outputs=\"text\",\n",
    "    title=\"Whisper Base Vietnamese\",\n",
    "    description=\"Realtime demo for Vietnamese speech recognition using a fine-tuned Whisper base model.\",\n",
    ")\n",
    "\n",
    "iface.launch()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "DUY",
   "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": "3.9.17"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}