TensorRT-LLM-Bloom-Example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "ac316c9b-6bd4-44bc-92ca-8455107156f2",
   "metadata": {},
   "source": [
    "# Getting started with TensorRT-LLM\n",
    "\n",
    "In this notebook we are optimizing Bloom 560M for inference using TensorRT-LLM. <br>\n",
    "We compare inference speed results from the **baseline** model (Huggingface), the **optimized** model, <br>\n",
    "and the optimized model after **INT8-quantization**.\n",
    "\n",
    "***Source:***\n",
    "https://github.com/NVIDIA/TensorRT-LLM/tree/main/examples/bloom"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "45885a26-ad28-473d-83d8-5c45bfc176d2",
   "metadata": {},
   "source": [
    "### Installing the NVIDIA Container Toolkit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "3c01a848-2619-4d7e-ad0a-d48529837c1d",
   "metadata": {},
   "outputs": [],
   "source": [
    "!curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \\\n",
    "  && curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \\\n",
    "    sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \\\n",
    "    sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list\n",
    "\n",
    "!apt-get update"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9d9e37e9-ef50-4e81-bab2-9c3ad4a94586",
   "metadata": {},
   "source": [
    "### Installing TensorRT-LLM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1b6333ce-4c4f-4fe9-b08f-c6db2242e00f",
   "metadata": {},
   "outputs": [],
   "source": [
    "!git clone https://github.com/NVIDIA/TensorRT-LLM/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "7f3f644f-c715-4262-8adc-4f67b9695566",
   "metadata": {},
   "outputs": [],
   "source": [
    "!apt-get update && apt-get -y install python3.10 python3-pip openmpi-bin libopenmpi-dev\n",
    "!pip3 install tensorrt_llm -U --pre --extra-index-url https://pypi.nvidia.com"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d029d061-72a2-4a9d-979e-0036324b4220",
   "metadata": {},
   "source": [
    "### Downloading BLOOM"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "846f303e-ef4d-42d0-8cd6-a7ec1e7ff677",
   "metadata": {},
   "source": [
    "BLOOM is an autoregressive Large Language Model (LLM), trained to continue text from a prompt on vast amounts of text data using industrial-scale computational resources. As such, it is able to output coherent text in 46 languages and 13 programming languages that is hardly distinguishable from text written by humans. BLOOM can also be instructed to perform text tasks it hasn't been explicitly trained for, by casting them as text generation tasks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "34dbc0ca-1cdb-46dc-82b2-c3e23ba9b18e",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install -r TensorRT-LLM/examples/bloom/requirements.txt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "25f9c485-9bbc-4dea-9d2d-240ac8f5ced2",
   "metadata": {},
   "outputs": [],
   "source": [
    "BLOOM_PATH=\"TensorRT-LLM/examples/bloom\"\n",
    "\n",
    "!rm -rf $BLOOM_PATH/560M\n",
    "!mkdir -p $BLOOM_PATH/560M && git clone https://huggingface.co/bigscience/bloom-560m $BLOOM_PATH/560M"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e0ab5234-a53e-4bdf-beb6-807dff0df9db",
   "metadata": {},
   "source": [
    "### Converting and Building"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "id": "c25e4c3a-94a4-40d7-8046-80b6cdfd8cba",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[TensorRT-LLM] TensorRT-LLM version: 0.9.0.dev2024030500\n",
      "0.9.0.dev2024030500\n",
      "[03/06/2024-20:31:52] [TRT-LLM] [I] Convert by using checkpoint\n",
      "[03/06/2024-20:31:52] [TRT-LLM] [I] Loading weights from HF BLOOM...\n",
      "Rank [0] Loading weights: 100%|███████████████████| 1/1 [00:05<00:00,  5.40s/it]\n",
      "[03/06/2024-20:31:58] [TRT-LLM] [I] Weights loaded. Total time: 00:00:05\n",
      "Total time of converting checkpoints: 00:00:26\n"
     ]
    }
   ],
   "source": [
    "# Convert checkpoint from HF to TRT-LLM format\n",
    "!python3 $BLOOM_PATH/convert_checkpoint.py --model_dir $BLOOM_PATH/560M/ \\\n",
    "                --dtype float16 \\\n",
    "                --output_dir $BLOOM_PATH/560M/trt_ckpt/fp16/1-gpu/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "653d0c5c-63ab-43ae-9378-5247b08dc4dd",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Build TensorRT-LLM model from checkpoint\n",
    "!trtllm-build --checkpoint_dir $BLOOM_PATH/560M/trt_ckpt/fp16/1-gpu/ \\\n",
    "                --gemm_plugin float16 \\\n",
    "                --output_dir $BLOOM_PATH/560M/trt_engines/fp16/1-gpu/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "beacb68b-10e3-4a60-b190-f9e0dd423dc1",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Build the BLOOM 560M using a single GPU and apply INT8 weight-only quantization.\n",
    "!python3 $BLOOM_PATH/convert_checkpoint.py --model_dir $BLOOM_PATH/560M \\\n",
    "                --dtype float16 \\\n",
    "                --use_weight_only \\\n",
    "                --output_dir $BLOOM_PATH/560M/trt_ckpt/int8_weight_only/1-gpu/\n",
    "\n",
    "!trtllm-build --checkpoint_dir $BLOOM_PATH/560M/trt_ckpt/int8_weight_only/1-gpu/ \\\n",
    "                --gemm_plugin float16 \\\n",
    "                --output_dir $BLOOM_PATH/560M/trt_engines/int8_weight_only/1-gpu/"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "96d19e11-3181-43af-ac1e-1d9a3ce99aab",
   "metadata": {},
   "source": [
    "# Benchmarking"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6b1bb3e0-67f1-4e61-b60b-bf329b273f9e",
   "metadata": {},
   "source": [
    "Comparing execution time and **ROUGE** metrics on a **Summarization Task**.\n",
    "\n",
    "**ROUGE** stands for **Recall-Oriented Understudy for Gisting Evaluation**.<br />\n",
    "It is a set of metrics used to evaluate the quality of summaries by comparing them to one or more reference summaries.<br />\n",
    "\n",
    "Since we are using NVIDIA's Python scripts for benchmarking, we need to capture **STDOUT** and parse the results from there."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "id": "0b28c0ad-eaa6-42cc-a9d1-785e60936d43",
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture bloom_hf_results\n",
    "# Huggingface\n",
    "!time python3 $BLOOM_PATH/../summarize.py --test_hf \\\n",
    "                       --hf_model_dir $BLOOM_PATH/560M/ \\\n",
    "                       --data_type fp16 \\\n",
    "                       --engine_dir $BLOOM_PATH/560M/trt_engines/fp16/1-gpu/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "id": "85de27d0-0d75-4ef1-a0da-5e17608e99fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture bloom_trt_results\n",
    "# TensorRT-LLM\n",
    "!time python3 $BLOOM_PATH/../summarize.py --test_trt_llm \\\n",
    "                       --hf_model_dir $BLOOM_PATH/560M/ \\\n",
    "                       --data_type fp16 \\\n",
    "                       --engine_dir $BLOOM_PATH/560M/trt_engines/fp16/1-gpu/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "id": "14d8d640-cd32-48f1-8d46-612871d53ca4",
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture bloom_int8_results\n",
    "# TensorRT-LLM (INT8)\n",
    "!time python3 $BLOOM_PATH/../summarize.py --test_trt_llm \\\n",
    "                       --hf_model_dir $BLOOM_PATH/560M/ \\\n",
    "                       --data_type fp16 \\\n",
    "                       --engine_dir $BLOOM_PATH/560M/trt_engines/int8_weight_only/1-gpu/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f37fa8c1-83b4-49cf-861e-20a382625faa",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Example output captured\n",
    "print(bloom_int8_results())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "54300987-e179-4b03-bb3c-39e9c2a80299",
   "metadata": {},
   "source": [
    "# Visualizing the results"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "56649eab-8b19-410a-8a73-573b13d553bf",
   "metadata": {},
   "source": [
    "#### Parse results from STDOUT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "id": "31223187-e3e7-4d1c-a867-d7ca4ebaa500",
   "metadata": {},
   "outputs": [],
   "source": [
    "import re\n",
    "\n",
    "def parse_output(output):\n",
    "    # Extracting execution time\n",
    "    time_match = re.search(r'real\\s+(\\d+)m([\\d.]+)s', output.stdout)\n",
    "    if time_match:\n",
    "        minutes = int(time_match.group(1))\n",
    "        seconds = float(time_match.group(2))\n",
    "        exec_time = minutes * 60 + seconds\n",
    "    else:\n",
    "        exec_time = None\n",
    "    \n",
    "    rouge_scores = re.findall(r'rouge[12Lsum]+ : ([\\d.]+)', output.stdout)\n",
    "    rouge_scores = [float(score) for score in rouge_scores] if rouge_scores else []\n",
    "    \n",
    "    latency_match = re.search(r'total latency: ([\\d.]+) sec', output.stdout)\n",
    "    latency = float(latency_match.group(1)) if latency_match else None\n",
    "    \n",
    "    tokens_match = re.search(r'total output tokens: (\\d+)', output.stdout)\n",
    "    total_tokens = int(tokens_match.group(1)) if tokens_match else None\n",
    "    \n",
    "    tokens_per_sec_match = re.search(r'tokens per second: ([\\d.]+)', output.stdout)\n",
    "    tokens_per_sec = float(tokens_per_sec_match.group(1)) if tokens_per_sec_match else None\n",
    "    \n",
    "    return exec_time, rouge_scores, latency, total_tokens, tokens_per_sec"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "id": "0d11d69a-36cf-4568-bcee-cd703823d2ae",
   "metadata": {},
   "outputs": [],
   "source": [
    "# For bloom_hf_results\n",
    "bloom_hf_exec_time, bloom_hf_rouge, _, _, _ = parse_output(bloom_hf_results)\n",
    "\n",
    "# Only TensorRT-LLM optimized models have Tokens/s, latency, and total tokens\n",
    "bloom_trt_exec_time, bloom_trt_rouge, bloom_trt_latency, bloom_trt_tokens, bloom_trt_tokens_per_sec = parse_output(bloom_trt_results)\n",
    "bloom_int8_exec_time, bloom_int8_rouge, bloom_int8_latency, bloom_int8_tokens, bloom_int8_tokens_per_sec = parse_output(bloom_int8_results)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "id": "a7429695-0d60-4dbd-bb54-21f3b88ae27f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'Model')"
      ]
     },
     "execution_count": 144,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# Execution times and Rouge Scores\n",
    "exec_times = [bloom_hf_exec_time, bloom_trt_exec_time, bloom_int8_exec_time]\n",
    "models = ['HF', 'TRT', 'INT8']\n",
    "\n",
    "plt.figure(figsize=(15, 5))\n",
    "\n",
    "# Plotting execution times\n",
    "plt.subplot(1, 3, 1)\n",
    "plt.bar(models, exec_times, color=['blue', 'green', 'red'])\n",
    "plt.title('Time Comparison (lower is better)')\n",
    "plt.ylabel('Execution Time (seconds)')\n",
    "plt.xlabel('Model')\n",
    "\n",
    "# Plotting ROUGE-1 scores\n",
    "rouge1_scores = [bloom_hf_rouge[0], bloom_trt_rouge[0], bloom_int8_rouge[0]]\n",
    "plt.subplot(1, 3, 2)\n",
    "plt.bar(models, rouge1_scores, color=['blue', 'green', 'red'])\n",
    "plt.title('ROUGE-1 (higher is better)')\n",
    "plt.ylabel('ROUGE-1 Score')\n",
    "plt.xlabel('Model')\n",
    "\n",
    "# Plotting ROUGE-Lsum scores\n",
    "rouge_lsum_scores = [bloom_hf_rouge[3], bloom_trt_rouge[3], bloom_int8_rouge[3]]\n",
    "plt.subplot(1, 3, 3)\n",
    "plt.bar(models, rouge_lsum_scores, color=['blue', 'green', 'red'])\n",
    "plt.title('ROUGE-Lsum (higher is better)')\n",
    "plt.ylabel('ROUGE-Lsum Score')\n",
    "plt.xlabel('Model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "id": "ecafe57c-3351-4e87-b0d2-6f78851e4900",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'Model')"
      ]
     },
     "execution_count": 145,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Latency and Tokens per Sec\n",
    "latencies = [bloom_trt_latency, bloom_int8_latency]\n",
    "tps = [bloom_trt_tokens_per_sec, bloom_int8_tokens_per_sec]\n",
    "models = ['TRT', 'INT8']\n",
    "\n",
    "plt.figure(figsize=(10, 5))\n",
    "\n",
    "# Plotting latency times\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.bar(models, latencies, color=['green', 'red'])\n",
    "plt.title('Latency (lower is better)')\n",
    "plt.ylabel('Time (seconds)')\n",
    "plt.xlabel('Model')\n",
    "\n",
    "# Plotting tokens per seconds\n",
    "rouge1_scores = tps\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.bar(models, rouge1_scores, color=['green', 'red'])\n",
    "plt.title('Tokens per second (higher is better)')\n",
    "plt.ylabel('Tokens/sec')\n",
    "plt.xlabel('Model')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}