foundation_model_notebooks

00_check_cpu_cuda_mps.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "deb52a72-f025-4b14-8f57-a8be049fb76c",
   "metadata": {},
   "source": [
    "#### Check current device availability"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "28b5f9e3-2c3b-4807-a18a-123f801b9b92",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch:[2.10.0]\n"
     ]
    }
   ],
   "source": [
    "import time\n",
    "import torch\n",
    "print (\"torch:[%s]\"%(torch.__version__))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "493a7567-45cf-49b2-a834-0ecc26d667cb",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ready.\n"
     ]
    }
   ],
   "source": [
    "def bench_matmul(dev,n=2048,warmup=10,iters=50):\n",
    "    dtype = torch.float32 if dev==\"cpu\" else torch.float16\n",
    "\n",
    "    a = torch.randn((n,n),device=dev,dtype=dtype)\n",
    "    b = torch.randn((n,n),device=dev,dtype=dtype)\n",
    "\n",
    "    for _ in range(warmup):\n",
    "        _ = a @ b\n",
    "    if dev==\"cuda\":\n",
    "        torch.cuda.synchronize()\n",
    "    elif dev==\"mps\":\n",
    "        torch.mps.synchronize()\n",
    "\n",
    "    t0 = time.time()\n",
    "    for _ in range(iters):\n",
    "        _ = a @ b\n",
    "    if dev==\"cuda\":\n",
    "        torch.cuda.synchronize()\n",
    "    elif dev==\"mps\":\n",
    "        torch.mps.synchronize()\n",
    "    t1 = time.time()\n",
    "\n",
    "    dt     = (t1-t0)/float(iters)\n",
    "    tflops = (2.0*(n**3)/dt)/1e12\n",
    "    return dt,tflops\n",
    "\n",
    "print (\"Ready.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "6019d7d5-5f02-47bc-988a-81fa23511bb8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cpu:[True]\n",
      "cuda:[False]\n",
      "mps:[True]\n",
      "mps_is_built:[True]\n"
     ]
    }
   ],
   "source": [
    "# Check availability\n",
    "cpu_ok  = True\n",
    "cuda_ok = torch.cuda.is_available()\n",
    "mps_ok  = (hasattr(torch.backends,\"mps\") and torch.backends.mps.is_available())\n",
    "\n",
    "print(\"cpu:[%s]\"%cpu_ok)\n",
    "print(\"cuda:[%s]\"%cuda_ok)\n",
    "print(\"mps:[%s]\"%mps_ok)\n",
    "\n",
    "if cuda_ok:\n",
    "    print(\"cuda_device_count:[%d]\"%torch.cuda.device_count())\n",
    "    print(\"cuda_device_name_0:[%s]\"%torch.cuda.get_device_name(0))\n",
    "\n",
    "if mps_ok:\n",
    "    print(\"mps_is_built:[%s]\"%torch.backends.mps.is_built())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "be217dff-267a-4c7f-810a-3b38065e3a4e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "n = [2048]\n",
      "FLOPs per matmul ≈ 2*n^3 = [17179869184]\n",
      "TFLOPs ≈ (FLOPs / avg_time) / 1e12\n",
      "[cpu]  avg=[0.005192]s  approx=[3.309] TFLOPs\n",
      "[mps]  avg=[0.001160]s  approx=[14.805] TFLOPs\n"
     ]
    }
   ],
   "source": [
    "n = 2048\n",
    "flops = 2*(n**3)\n",
    "print(\"n = [%d]\"%n)\n",
    "print(\"FLOPs per matmul ≈ 2*n^3 = [%d]\"%flops)\n",
    "print(\"TFLOPs ≈ (FLOPs / avg_time) / 1e12\")\n",
    "\n",
    "if cpu_ok:\n",
    "    dt,tflops = bench_matmul(\"cpu\",n=n,warmup=3,iters=10)\n",
    "    print(\"[cpu]  avg=[%.6f]s  approx=[%.3f] TFLOPs\" % (dt,tflops))\n",
    "if cuda_ok:\n",
    "    dt,tflops = bench_matmul(\"cuda\",n=n,warmup=10,iters=50)\n",
    "    print(\"[cuda] avg=[%.6f]s  approx=[%.3f] TFLOPs\" % (dt,tflops))\n",
    "if mps_ok:\n",
    "    dt,tflops = bench_matmul(\"mps\",n=n,warmup=10,iters=50)\n",
    "    print(\"[mps]  avg=[%.6f]s  approx=[%.3f] TFLOPs\" % (dt,tflops))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f8af04a6-71e7-4260-b42e-e0ecebabef42",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}

01_hf_sam2.ipynb

02_hf_owlv2.ipynb

03_hf_owlv2_sam2.ipynb

04_hf_qwen3_vl.ipynb

10_openai_gpt5_usage.ipynb