No public description

notebooks/community/model_garden/jax_vision_transformer_local_docker.ipynb

@@ -0,0 +1,162 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "3c19749c-4780-47e7-9da1-c4a86aa227dc",
+   "metadata": {},
+   "source": [
+    "# Local docker run for JAX VIT training\n",
+    "\n",
+    "This notebook shows local docker run for JAX VIT training.\n",
+    "This notebook uses a workbench with TensorFlow 2.11 and 8 v100 GPUs.\n",
+    "You also need to upload the 'train_vit_gpu.Dockerfile' to the home directory for workbench."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a3df6e8a-68c7-40a2-9b24-0ed7acdeffc3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Build training docker\n",
+    "\n",
+    "project=\"cloud-nas-260507\"\n",
+    "image_tag=\"jax-vit-train-gpu-lavrai-test:latest\"\n",
+    "train_docker_uri=\"gcr.io/{}/{}\".format(project, image_tag)\n",
+    "\n",
+    "!docker build -f train_vit_gpu.Dockerfile . -t {image_tag}\n",
+    "\n",
+    "!docker tag {image_tag} {train_docker_uri}\n",
+    "\n",
+    "!docker push {train_docker_uri}"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "10978e1a-8cf2-4a07-8b0c-f0e45a88bf53",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Docker arguments.\n",
+    "workdir='tmp'\n",
+    "docker_args_list=[\n",
+    "    '--config', 'vit_jax/configs/augreg.py:R_Ti_16',\n",
+    "    '--config.dataset', 'tf_flowers',\n",
+    "    '--config.pp.train', 'train[:90%]',\n",
+    "    '--config.pp.test', 'train[90%:]',\n",
+    "    '--config.batch_eval', '120',\n",
+    "    '--config.base_lr', '0.01',\n",
+    "    '--config.shuffle_buffer', '1000',\n",
+    "    '--config.total_steps', '100',\n",
+    "    '--config.warmup_steps', '10',\n",
+    "    '--config.accum_steps', '0',  # Not needed with R+Ti/16 model.\n",
+    "    '--config.pp.crop', '224',\n",
+    "    '--workdir', f'{workdir}',\n",
+    "                 ]\n",
+    "print(docker_args_list)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "b36477a6-f083-4098-8745-aebcc6e74098",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Utility functions.\n",
+    "\n",
+    "import io\n",
+    "import subprocess\n",
+    "import sys\n",
+    "\n",
+    "def run_command_with_stdout(cmd, job_log_file=None, error_message=\"\"):\n",
+    "  \"\"\"Runs the command and stream the command outputs.\"\"\"\n",
+    "  if job_log_file is None:\n",
+    "    job_log_file = sys.stdout\n",
+    "  buf = io.StringIO()\n",
+    "  ret_code = None\n",
+    "\n",
+    "  with subprocess.Popen(\n",
+    "      cmd,\n",
+    "      stdin=subprocess.PIPE,\n",
+    "      stdout=subprocess.PIPE,\n",
+    "      stderr=subprocess.STDOUT,\n",
+    "      universal_newlines=False,\n",
+    "  ) as p:\n",
+    "    out = io.TextIOWrapper(p.stdout, newline=\"\")\n",
+    "\n",
+    "    for line in out:\n",
+    "      buf.write(line)\n",
+    "      job_log_file.write(line)\n",
+    "      job_log_file.flush()\n",
+    "\n",
+    "    # flush to force the contents to display.\n",
+    "    job_log_file.flush()\n",
+    "\n",
+    "    while p.poll() is None:\n",
+    "      # Process hasn't exited yet, let's wait some\n",
+    "      time.sleep(0.5)\n",
+    "\n",
+    "    ret_code = p.returncode\n",
+    "    p.stdout.close()\n",
+    "\n",
+    "  if ret_code:\n",
+    "    raise RuntimeError(\n",
+    "        \"Error: {} with return code {}\".format(error_message, ret_code)\n",
+    "    )\n",
+    "  return buf.getvalue(), ret_code\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7d559395-c1d5-4ccd-8066-054eac3ad2d2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Run local training.\n",
+    "cmd = ([\"nvidia-docker\", \"run\"] + [\"-t\", train_docker_uri] + docker_args_list)\n",
+    "run_command_with_stdout(\n",
+    "  cmd, error_message=\"Failed to run docker locally\"\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "88367395-5608-4262-82d9-e94c55f15903",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "environment": {
+   "kernel": "python3",
+   "name": "tf2-gpu.2-11.m108",
+   "type": "gcloud",
+   "uri": "gcr.io/deeplearning-platform-release/tf2-gpu.2-11:m108"
+  },
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.7.12"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}