hindi_stt_issue.ipynb

hindi_stt_issue.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/gist/Yugsolanki/eea86a762d519449a6e254cc8ef84c7e/hindi_stt_issue.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Installation & Extraction"
      ],
      "metadata": {
        "id": "bqGHCtFwgdIt"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!uv install gdown\n",
        "\n",
        "import gdown\n",
        "\n",
        "url = 'https://drive.google.com/uc?export=download&id=1tNhIbvKxc__9x9xnoI-AsOh2qMxYUwM8'\n",
        "output = 'colab_data.zip'\n",
        "gdown.download(url, output, quiet=False)\n",
        "\n",
        "!unzip colab_data.zip"
      ],
      "metadata": {
        "id": "G4MiTp2if4sT"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "NDCbYfESJ0kM"
      },
      "outputs": [],
      "source": [
        "!uv pip install pytorch-lightning nemo-toolkit[asr] \"cuda-python==12.8.0\" \"numba-cuda==0.4.0\" \"numpy==1.26.4\" megatron megatron-core"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "---"
      ],
      "metadata": {
        "id": "ZPUF8whmgsAt"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Training"
      ],
      "metadata": {
        "id": "A9J1XCrDgsf4"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "pwOn35rwKx8w"
      },
      "outputs": [],
      "source": [
        "# Training the model (Hindi streaming FastConformer Hybrid RNNT+CTC with SPE-Unigram)\n",
        "import lightning.pytorch as pl\n",
        "from omegaconf import OmegaConf\n",
        "\n",
        "from nemo.collections.asr.models import EncDecHybridRNNTCTCBPEModel\n",
        "from nemo.core.config import hydra_runner\n",
        "from nemo.utils import logging\n",
        "from nemo.utils.exp_manager import exp_manager\n",
        "from nemo.utils.trainer_utils import resolve_trainer_cfg\n",
        "from nemo.utils.exp_manager import EarlyStopping, ModelCheckpoint"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "gFhaWSiXK93P"
      },
      "outputs": [],
      "source": [
        "def main(cfg):\n",
        "    logging.info(f'Hydra config:\\n{OmegaConf.to_yaml(cfg)}')\n",
        "\n",
        "    try:\n",
        "        early_stopping_callback = EarlyStopping(\n",
        "            monitor=\"val_wer\",\n",
        "            mode=\"min\",\n",
        "            patience=10,\n",
        "            min_delta=0.0001,\n",
        "            verbose=True\n",
        "        )\n",
        "\n",
        "        checkpoint_callback = ModelCheckpoint(\n",
        "            monitor=\"val_wer\",\n",
        "            mode=\"min\",\n",
        "            save_top_k=5,\n",
        "            every_n_train_steps=2000,\n",
        "            filename=\"${name}--{reduced_train_loss:.2f}-{step}\"\n",
        "        )\n",
        "\n",
        "        trainer = pl.Trainer(**resolve_trainer_cfg(cfg.trainer),callbacks=[early_stopping_callback, checkpoint_callback])\n",
        "        exp_manager(trainer, cfg.get(\"exp_manager\", None))\n",
        "        asr_model = EncDecHybridRNNTCTCBPEModel.from_config_dict(cfg.model, trainer=trainer)\n",
        "\n",
        "        asr_model.change_vocabulary(\n",
        "            new_tokenizer_dir=cfg.model.tokenizer.dir,\n",
        "            new_tokenizer_type=cfg.model.tokenizer.type\n",
        "        )\n",
        "\n",
        "        logging.info(\"Model configuration AFTER all updates:\")\n",
        "        logging.info(asr_model.cfg)\n",
        "\n",
        "        # Initialize the weights of the model from another model, if provided via config\n",
        "        asr_model.maybe_init_from_pretrained_checkpoint(cfg)\n",
        "\n",
        "        trainer.fit(asr_model)\n",
        "\n",
        "        if hasattr(cfg.model, 'test_ds') and cfg.model.test_ds.manifest_filepath is not None:\n",
        "            if asr_model.prepare_test(trainer):\n",
        "                trainer.test(asr_model)\n",
        "\n",
        "        if not cfg.exp_manager.get(\"create_checkpoint_callback\", True):\n",
        "            final_save_path = os.path.join(\n",
        "                cfg.exp_manager.exp_dir,\n",
        "                cfg.exp_manager.name,\n",
        "                \"final_model.nemo\"\n",
        "            )\n",
        "            logging.info(f\"Saving final model to: {final_save_path}\")\n",
        "            asr_model.save_to(final_save_path)\n",
        "    except KeyboardInterrupt:\n",
        "        logging.warning(\"Training interrupted by user (Ctrl+C)\")\n",
        "        logging.info(\"Attempting to save checkpoint before exit...\")\n",
        "        raise\n",
        "\n",
        "    except Exception as e:\n",
        "        logging.error(\"=\" * 80)\n",
        "        logging.error(f\"Training failed with error: {str(e)}\")\n",
        "        logging.error(\"=\" * 80)\n",
        "        raise"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hk-m_VLkNz-9"
      },
      "outputs": [],
      "source": [
        "from omegaconf import OmegaConf\n",
        "\n",
        "cfg = OmegaConf.load(\"/content/colab_data/config.yaml\")\n",
        "\n",
        "main(cfg)"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "gpuType": "T4",
      "provenance": [],
      "mount_file_id": "1_1IYEQ4UtTDwYDxzhMM3wQ0U5tT9gZqu",
      "authorship_tag": "ABX9TyP4soCKrbX8v6yuwVc3V8fw",
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}