projection.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "eb078c07-fa57-4cd1-92a9-0fb8dd273ca9",
   "metadata": {},
   "source": [
    "## Setup and Input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "2bda4040-9b62-49df-9179-61bfd83b2eef",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "sys.path.append(\"../projection\")\n",
    "from tqdm import tqdm\n",
    "from moe_projection import *\n",
    "\n",
    "\n",
    "hidden_size = 4096\n",
    "num_heads_q = 32\n",
    "num_heads_kv = 8\n",
    "intermediate_size = 14336\n",
    "mlp_with_gate = True\n",
    "num_experts = 8\n",
    "num_layers = 32\n",
    "\n",
    "dtype_list = [\"bf16\", \"fp8\"]\n",
    "# in_out_token_list = [{\"in\": 128, \"out\": 128}, {\"in\": 1024, \"out\": 1024}, {\"in\": 1, \"out\": 2048}, {\"in\": 32000, \"out\": 512}]\n",
    "in_out_token_list = [{\"in\": 128, \"out\": 128}]\n",
    "batchsize_list = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]\n",
    "\n",
    "projection_dict = {\"prefill\": {}, \"decode\": {}}\n",
    "analysis_dict = {\"prefill\": [], \"decode\": []}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9c8126f1-a6e3-4135-85e8-cb11611ca8ca",
   "metadata": {},
   "source": [
    "## Prefill Projection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "e82bfd07-acea-43c5-83bc-305d3efc5d3f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "projection prefill with dtype[bf16], device [Gaudi2C_BF16] with seq_len: [{'in': 128, 'out': 128}] and bs [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 26732.34it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "done!\n",
      "\n",
      "projection prefill with dtype[fp8], device [Gaudi2C_FP8] with seq_len: [{'in': 128, 'out': 128}] and bs [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 28591.03it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "done!\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "for dtype in dtype_list:\n",
    "    device = type2devices[dtype]\n",
    "    num_bytes = type2bytes[dtype]\n",
    "    bw = device_bw_tops[device][0]\n",
    "    tops = device_bw_tops[device][1]\n",
    "    tops_tpc = device_bw_tops[device][2]\n",
    "\n",
    "    # prefill\n",
    "    print(\n",
    "        f\"projection prefill with dtype[{dtype}], device [{device}] with seq_len: {in_out_token_list} and bs {batchsize_list}...\")\n",
    "    projection_dict[\"prefill\"][dtype] = []\n",
    "    prefill_projection = [item_list]\n",
    "    prefill_layer_analysis = dict()\n",
    "    for in_out in in_out_token_list:\n",
    "        for bs in tqdm(batchsize_list):\n",
    "            prefill_layer_analysis[bs] = [layer_analysis_list]\n",
    "            model_config = Config(batch_size=bs, seq_len_q=in_out[\"in\"], seq_len_kv=in_out[\"in\"], hidden_size=hidden_size, num_heads_q=num_heads_q,\n",
    "                                  num_heads_kv=num_heads_kv, intermediate_size=intermediate_size, is_decoding=False, num_bytes=num_bytes,\n",
    "                                  bw=bw, tops=tops, tops_tpc=tops_tpc, with_gate=mlp_with_gate, num_experts=num_experts, num_layers=num_layers)\n",
    "            runtime_decoder, single_layer_items = proj_decoder(\n",
    "                model_config)\n",
    "            prefill_projection.append([device, model_config.hidden_size, model_config.num_heads_q, model_config.num_heads_kv,\n",
    "                                       model_config.intermediate_size, model_config.is_decoding, model_config.num_experts,\n",
    "                                       model_config.num_layers, in_out[\"in\"], in_out[\"out\"], dtype, bs, round(runtime_decoder, 2),\n",
    "                                       round(1/runtime_decoder * model_config.batch_size, 2)])\n",
    "            prefill_layer_analysis[bs].append(\n",
    "                [in_out[\"in\"], in_out[\"out\"], dtype, bs, single_layer_items[\"qkvo\"][\"name\"], round(single_layer_items[\"qkvo\"][\"#ops\"]/1e9, 2),\n",
    "                 round(single_layer_items[\"qkvo\"][\"#mem\"]/1024/1024/1024,\n",
    "                       2), round(single_layer_items[\"qkvo\"][\"tops_roofline\"]/1e12, 2),\n",
    "                 round(single_layer_items[\"qkvo\"][\"math_ai\"], 2), single_layer_items[\"qkvo\"][\"bound\"]])\n",
    "            for item in single_layer_items[\"attn\"]:\n",
    "                prefill_layer_analysis[bs].append(\n",
    "                    [in_out[\"in\"], in_out[\"out\"], dtype, bs, item[\"name\"], round(item[\"#ops\"]/1e9, 2), round(item[\"#mem\"]/1024/1024/1024, 2),\n",
    "                     round(item[\"tops_roofline\"]/1e12, 2), round(item[\"math_ai\"], 2), item[\"bound\"]])\n",
    "            for item in single_layer_items[\"moe\"]:\n",
    "                prefill_layer_analysis[bs].append(\n",
    "                    [in_out[\"in\"], in_out[\"out\"], dtype, bs, item[\"name\"], round(item[\"#ops\"]/1e9, 2), round(item[\"#mem\"]/1024/1024/1024, 2),\n",
    "                     round(item[\"tops_roofline\"]/1e12, 2), round(item[\"math_ai\"], 2), item[\"bound\"]])\n",
    "    print(\"done!\\n\")\n",
    "    # print(tabulate(prefill_projection))\n",
    "    # for bs in batchsize_list:\n",
    "    #     print(tabulate(prefill_layer_analysis[bs]))\n",
    "    projection_dict[\"prefill\"][dtype].append(prefill_projection)\n",
    "    analysis_dict[\"prefill\"].append(prefill_layer_analysis)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e6ca6078-9c17-45d4-80dd-3e41385a884e",
   "metadata": {},
   "source": [
    "## Decode Projection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "17ac5309-0a06-42cf-95e7-fd46498f914e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "projection decoding with dtype[bf16], device [Gaudi2C_BF16] with seq_len: [{'in': 128, 'out': 128}] and bs [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 28767.52it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "done!\n",
      "projection decoding with dtype[fp8], device [Gaudi2C_FP8] with seq_len: [{'in': 128, 'out': 128}] and bs [1, 2, 4, 8, 16, 32, 64, 128, 256, 512]...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 30750.03it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "done!\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "for dtype in dtype_list:\n",
    "    device = type2devices[dtype]\n",
    "    num_bytes = type2bytes[dtype]\n",
    "    bw = device_bw_tops[device][0]\n",
    "    tops = device_bw_tops[device][1]\n",
    "    tops_tpc = device_bw_tops[device][2]\n",
    "\n",
    "    # decode\n",
    "    print(\n",
    "        f\"projection decoding with dtype[{dtype}], device [{device}] with seq_len: {in_out_token_list} and bs {batchsize_list}...\")\n",
    "    projection_dict[\"decode\"][dtype] = []\n",
    "    decoding_projection = [item_list]\n",
    "    decoding_layer_analysis = dict()\n",
    "    for in_out in in_out_token_list:\n",
    "        for bs in tqdm(batchsize_list):\n",
    "            decoding_layer_analysis[bs] = [layer_analysis_list]\n",
    "            model_config = Config(batch_size=bs, seq_len_q=1, seq_len_kv=in_out[\"out\"], hidden_size=hidden_size, num_heads_q=num_heads_q,\n",
    "                                  num_heads_kv=num_heads_kv, intermediate_size=intermediate_size, is_decoding=True, num_bytes=num_bytes,\n",
    "                                  bw=bw, tops=tops, tops_tpc=tops_tpc, with_gate=mlp_with_gate, num_experts=num_experts, num_layers=num_layers)\n",
    "            runtime_decoder, single_layer_items = proj_decoder(\n",
    "                model_config)\n",
    "            decoding_projection.append([device, model_config.hidden_size, model_config.num_heads_q, model_config.num_heads_kv,\n",
    "                                        model_config.intermediate_size, model_config.is_decoding, model_config.num_experts,\n",
    "                                        model_config.num_layers, in_out[\"in\"], in_out[\"out\"], dtype, bs, round(runtime_decoder, 2),\n",
    "                                        round(1/runtime_decoder * model_config.batch_size, 2)])\n",
    "            decoding_layer_analysis[bs].append(\n",
    "                [in_out[\"in\"], in_out[\"out\"], dtype, bs, single_layer_items[\"qkvo\"][\"name\"], round(single_layer_items[\"qkvo\"][\"#ops\"]/1e9, 2),\n",
    "                 round(single_layer_items[\"qkvo\"][\"#mem\"]/1024/1024/1024,\n",
    "                       2), round(single_layer_items[\"qkvo\"][\"tops_roofline\"]/1e12, 2),\n",
    "                 round(single_layer_items[\"qkvo\"][\"math_ai\"], 2), single_layer_items[\"qkvo\"][\"bound\"]]\n",
    "            )\n",
    "            for item in single_layer_items[\"attn\"]:\n",
    "                decoding_layer_analysis[bs].append(\n",
    "                    [in_out[\"in\"], in_out[\"out\"], dtype, bs, item[\"name\"], round(item[\"#ops\"]/1e9, 2), round(item[\"#mem\"]/1024/1024/1024, 2),\n",
    "                     round(item[\"tops_roofline\"]/1e12, 2), round(item[\"math_ai\"], 2), item[\"bound\"]])\n",
    "            for item in single_layer_items[\"moe\"]:\n",
    "                decoding_layer_analysis[bs].append(\n",
    "                    [in_out[\"in\"], in_out[\"out\"], dtype, bs, item[\"name\"], round(item[\"#ops\"]/1e9, 2), round(item[\"#mem\"]/1024/1024/1024, 2),\n",
    "                     round(item[\"tops_roofline\"]/1e12, 2), round(item[\"math_ai\"], 2), item[\"bound\"]])\n",
    "    print(\"done!\")\n",
    "    # print(tabulate(decoding_projection))\n",
    "    # for bs in batchsize_list:\n",
    "    #     print(tabulate(decoding_layer_analysis[bs]))\n",
    "    projection_dict[\"decode\"][dtype].append(decoding_projection)\n",
    "    analysis_dict[\"decode\"].append(decoding_layer_analysis)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bc6fcd57-842a-485c-941b-c43910e89018",
   "metadata": {},
   "source": [
    "## Projection Result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "af4ae18b-6bf4-43a8-9bc7-7b1296fd2d65",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 2000x1000 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                                                       prefill                                                                        \n",
      "------------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "Device        HiddenSize  HeadsQ  HeadsKV  InterSize  Decoding  Experts  Layers  Input  Output  DType  BS   Latency(s)  Throughput(tokens/sec)\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   1    0.04        22.48\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   2    0.09        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   4    0.17        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   8    0.34        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   16   0.68        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   32   1.37        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   64   2.73        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   128  5.47        23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   256  10.93       23.42\n",
      "Gaudi2C_BF16  4096        32      8        14336      False     8        32      128    128     bf16   512  21.86       23.42\n",
      "------------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "                                                                       prefill                                                                        \n",
      "-----------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "Device       HiddenSize  HeadsQ  HeadsKV  InterSize  Decoding  Experts  Layers  Input  Output  DType  BS   Latency(s)  Throughput(tokens/sec)\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    1    0.02        44.48\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    2    0.04        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    4    0.09        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    8    0.17        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    16   0.35        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    32   0.69        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    64   1.38        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    128  2.76        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    256  5.53        46.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      False     8        32      128    128     fp8    512  11.06       46.31\n",
      "-----------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "                                                                        decode                                                                        \n",
      "------------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "Device        HiddenSize  HeadsQ  HeadsKV  InterSize  Decoding  Experts  Layers  Input  Output  DType  BS   Latency(s)  Throughput(tokens/sec)\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   1    0.04        23.68\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   2    0.04        47.34\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   4    0.04        94.6\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   8    0.04        188.74\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   16   0.04        375.69\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   32   0.04        744.29\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   64   0.04        1461.02\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   128  0.05        2817.71\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   256  0.09        2932.54\n",
      "Gaudi2C_BF16  4096        32      8        14336      True      8        32      128    128     bf16   512  0.17        2932.54\n",
      "------------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "                                                                        decode                                                                        \n",
      "-----------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n",
      "Device       HiddenSize  HeadsQ  HeadsKV  InterSize  Decoding  Experts  Layers  Input  Output  DType  BS   Latency(s)  Throughput(tokens/sec)\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    1    0.02        47.36\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    2    0.02        94.66\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    4    0.02        189.13\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    8    0.02        377.22\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    16   0.02        750.31\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    32   0.02        1484.37\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    64   0.02        2905.86\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    128  0.02        5575.51\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    256  0.04        5800.21\n",
      "Gaudi2C_FP8  4096        32      8        14336      True      8        32      128    128     fp8    512  0.09        5800.21\n",
      "-----------  ----------  ------  -------  ---------  --------  -------  ------  -----  ------  -----  ---  ----------  ----------------------\n"
     ]
    }
   ],
   "source": [
    "plot_projection(projection_dict, batchsize_list)\n",
    "print_projection(projection_dict)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c49cc1cc-0c51-4b63-b40c-b68dfb0d228f",
   "metadata": {},
   "source": [
    "## Analysis Result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "b21fd3a0-9dea-429f-9575-42ea69793983",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      1          qkvo_proj     17.18       0.13        269.85          120.47  memory\n",
      "128    128     bf16      1          q@k_T         0.13        0.0         127.43          56.89   memory\n",
      "128    128     bf16      1          softmax       0.34        0.0         22.0            160.0   compute\n",
      "128    128     bf16      1          score@v       0.13        0.0         127.43          56.89   memory\n",
      "128    128     bf16      1          mlp_up(w1)    15.03       0.11        275.64          123.06  memory\n",
      "128    128     bf16      1          mlp_down(w2)  15.03       0.11        275.64          123.06  memory\n",
      "128    128     bf16      1          mlp_gate(w3)  15.03       0.11        275.64          123.06  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       1          qkvo_proj     17.18       0.07        539.71          240.94  memory\n",
      "128    128     fp8       1          q@k_T         0.13        0.0         254.86          113.78  memory\n",
      "128    128     fp8       1          softmax       0.34        0.0         22.0            160.0   compute\n",
      "128    128     fp8       1          score@v       0.13        0.0         254.86          113.78  memory\n",
      "128    128     fp8       1          mlp_up(w1)    15.03       0.06        551.29          246.11  memory\n",
      "128    128     fp8       1          mlp_down(w2)  15.03       0.06        551.29          246.11  memory\n",
      "128    128     fp8       1          mlp_gate(w3)  15.03       0.06        551.29          246.11  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      2          qkvo_proj     34.36       0.14        287.0           227.56  compute\n",
      "128    128     bf16      2          q@k_T         0.27        0.0         127.43          56.89   memory\n",
      "128    128     bf16      2          softmax       0.67        0.0         22.0            160.0   compute\n",
      "128    128     bf16      2          score@v       0.27        0.0         127.43          56.89   memory\n",
      "128    128     bf16      2          mlp_up(w1)    30.06       0.12        287.0           236.96  compute\n",
      "128    128     bf16      2          mlp_down(w2)  30.06       0.12        287.0           236.96  compute\n",
      "128    128     bf16      2          mlp_gate(w3)  30.06       0.12        287.0           236.96  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       2          qkvo_proj     34.36       0.07        574.0           455.11  compute\n",
      "128    128     fp8       2          q@k_T         0.27        0.0         254.86          113.78  memory\n",
      "128    128     fp8       2          softmax       0.67        0.0         22.0            160.0   compute\n",
      "128    128     fp8       2          score@v       0.27        0.0         254.86          113.78  memory\n",
      "128    128     fp8       2          mlp_up(w1)    30.06       0.06        574.0           473.92  compute\n",
      "128    128     fp8       2          mlp_down(w2)  30.06       0.06        574.0           473.92  compute\n",
      "128    128     fp8       2          mlp_gate(w3)  30.06       0.06        574.0           473.92  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      4          qkvo_proj     68.72       0.16        287.0           409.6   compute\n",
      "128    128     bf16      4          q@k_T         0.54        0.01        127.43          56.89   memory\n",
      "128    128     bf16      4          softmax       1.34        0.01        22.0            160.0   compute\n",
      "128    128     bf16      4          score@v       0.54        0.01        127.43          56.89   memory\n",
      "128    128     bf16      4          mlp_up(w1)    60.13       0.13        287.0           441.11  compute\n",
      "128    128     bf16      4          mlp_down(w2)  60.13       0.13        287.0           441.11  compute\n",
      "128    128     bf16      4          mlp_gate(w3)  60.13       0.13        287.0           441.11  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       4          qkvo_proj     68.72       0.08        574.0           819.2   compute\n",
      "128    128     fp8       4          q@k_T         0.54        0.0         254.86          113.78  memory\n",
      "128    128     fp8       4          softmax       1.34        0.01        22.0            160.0   compute\n",
      "128    128     fp8       4          score@v       0.54        0.0         254.86          113.78  memory\n",
      "128    128     fp8       4          mlp_up(w1)    60.13       0.06        574.0           882.22  compute\n",
      "128    128     fp8       4          mlp_down(w2)  60.13       0.06        574.0           882.22  compute\n",
      "128    128     fp8       4          mlp_gate(w3)  60.13       0.06        574.0           882.22  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      8          qkvo_proj     137.44      0.19        287.0           682.67  compute\n",
      "128    128     bf16      8          q@k_T         1.07        0.02        127.43          56.89   memory\n",
      "128    128     bf16      8          softmax       2.68        0.02        22.0            160.0   compute\n",
      "128    128     bf16      8          score@v       1.07        0.02        127.43          56.89   memory\n",
      "128    128     bf16      8          mlp_up(w1)    120.26      0.14        287.0           774.92  compute\n",
      "128    128     bf16      8          mlp_down(w2)  120.26      0.14        287.0           774.92  compute\n",
      "128    128     bf16      8          mlp_gate(w3)  120.26      0.14        287.0           774.92  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       8          qkvo_proj     137.44      0.09        574.0           1365.33  compute\n",
      "128    128     fp8       8          q@k_T         1.07        0.01        254.86          113.78   memory\n",
      "128    128     fp8       8          softmax       2.68        0.02        22.0            160.0    compute\n",
      "128    128     fp8       8          score@v       1.07        0.01        254.86          113.78   memory\n",
      "128    128     fp8       8          mlp_up(w1)    120.26      0.07        574.0           1549.84  compute\n",
      "128    128     fp8       8          mlp_down(w2)  120.26      0.07        574.0           1549.84  compute\n",
      "128    128     fp8       8          mlp_gate(w3)  120.26      0.07        574.0           1549.84  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     bf16      16         qkvo_proj     274.88      0.25        287.0           1024.0   compute\n",
      "128    128     bf16      16         q@k_T         2.15        0.04        127.43          56.89    memory\n",
      "128    128     bf16      16         softmax       5.37        0.03        22.0            160.0    compute\n",
      "128    128     bf16      16         score@v       2.15        0.04        127.43          56.89    memory\n",
      "128    128     bf16      16         mlp_up(w1)    240.52      0.18        287.0           1246.61  compute\n",
      "128    128     bf16      16         mlp_down(w2)  240.52      0.18        287.0           1246.61  compute\n",
      "128    128     bf16      16         mlp_gate(w3)  240.52      0.18        287.0           1246.61  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       16         qkvo_proj     274.88      0.12        574.0           2048.0   compute\n",
      "128    128     fp8       16         q@k_T         2.15        0.02        254.86          113.78   memory\n",
      "128    128     fp8       16         softmax       5.37        0.03        22.0            160.0    compute\n",
      "128    128     fp8       16         score@v       2.15        0.02        254.86          113.78   memory\n",
      "128    128     fp8       16         mlp_up(w1)    240.52      0.09        574.0           2493.22  compute\n",
      "128    128     fp8       16         mlp_down(w2)  240.52      0.09        574.0           2493.22  compute\n",
      "128    128     fp8       16         mlp_gate(w3)  240.52      0.09        574.0           2493.22  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     bf16      32         qkvo_proj     549.76      0.38        287.0           1365.33  compute\n",
      "128    128     bf16      32         q@k_T         4.29        0.07        127.43          56.89    memory\n",
      "128    128     bf16      32         softmax       10.74       0.06        22.0            160.0    compute\n",
      "128    128     bf16      32         score@v       4.29        0.07        127.43          56.89    memory\n",
      "128    128     bf16      32         mlp_up(w1)    481.04      0.25        287.0           1792.0   compute\n",
      "128    128     bf16      32         mlp_down(w2)  481.04      0.25        287.0           1792.0   compute\n",
      "128    128     bf16      32         mlp_gate(w3)  481.04      0.25        287.0           1792.0   compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       32         qkvo_proj     549.76      0.19        574.0           2730.67  compute\n",
      "128    128     fp8       32         q@k_T         4.29        0.04        254.86          113.78   memory\n",
      "128    128     fp8       32         softmax       10.74       0.06        22.0            160.0    compute\n",
      "128    128     fp8       32         score@v       4.29        0.04        254.86          113.78   memory\n",
      "128    128     fp8       32         mlp_up(w1)    481.04      0.12        574.0           3584.0   compute\n",
      "128    128     fp8       32         mlp_down(w2)  481.04      0.12        574.0           3584.0   compute\n",
      "128    128     fp8       32         mlp_gate(w3)  481.04      0.12        574.0           3584.0   compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     bf16      64         qkvo_proj     1099.51     0.62        287.0           1638.4   compute\n",
      "128    128     bf16      64         q@k_T         8.59        0.14        127.43          56.89    memory\n",
      "128    128     bf16      64         softmax       21.47       0.12        22.0            160.0    compute\n",
      "128    128     bf16      64         score@v       8.59        0.14        127.43          56.89    memory\n",
      "128    128     bf16      64         mlp_up(w1)    962.07      0.39        287.0           2293.76  compute\n",
      "128    128     bf16      64         mlp_down(w2)  962.07      0.39        287.0           2293.76  compute\n",
      "128    128     bf16      64         mlp_gate(w3)  962.07      0.39        287.0           2293.76  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       64         qkvo_proj     1099.51     0.31        574.0           3276.8   compute\n",
      "128    128     fp8       64         q@k_T         8.59        0.07        254.86          113.78   memory\n",
      "128    128     fp8       64         softmax       21.47       0.12        22.0            160.0    compute\n",
      "128    128     fp8       64         score@v       8.59        0.07        254.86          113.78   memory\n",
      "128    128     fp8       64         mlp_up(w1)    962.07      0.2         574.0           4587.52  compute\n",
      "128    128     fp8       64         mlp_down(w2)  962.07      0.2         574.0           4587.52  compute\n",
      "128    128     fp8       64         mlp_gate(w3)  962.07      0.2         574.0           4587.52  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     bf16      128        qkvo_proj     2199.02     1.12        287.0           1820.44  compute\n",
      "128    128     bf16      128        q@k_T         17.18       0.28        127.43          56.89    memory\n",
      "128    128     bf16      128        softmax       42.95       0.25        22.0            160.0    compute\n",
      "128    128     bf16      128        score@v       17.18       0.28        127.43          56.89    memory\n",
      "128    128     bf16      128        mlp_up(w1)    1924.15     0.67        287.0           2667.16  compute\n",
      "128    128     bf16      128        mlp_down(w2)  1924.15     0.67        287.0           2667.16  compute\n",
      "128    128     bf16      128        mlp_gate(w3)  1924.15     0.67        287.0           2667.16  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       128        qkvo_proj     2199.02     0.56        574.0           3640.89  compute\n",
      "128    128     fp8       128        q@k_T         17.18       0.14        254.86          113.78   memory\n",
      "128    128     fp8       128        softmax       42.95       0.25        22.0            160.0    compute\n",
      "128    128     fp8       128        score@v       17.18       0.14        254.86          113.78   memory\n",
      "128    128     fp8       128        mlp_up(w1)    1924.15     0.34        574.0           5334.33  compute\n",
      "128    128     fp8       128        mlp_down(w2)  1924.15     0.34        574.0           5334.33  compute\n",
      "128    128     fp8       128        mlp_gate(w3)  1924.15     0.34        574.0           5334.33  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     bf16      256        qkvo_proj     4398.05     2.12        287.0           1927.53  compute\n",
      "128    128     bf16      256        q@k_T         34.36       0.56        127.43          56.89    memory\n",
      "128    128     bf16      256        softmax       85.9        0.5         22.0            160.0    compute\n",
      "128    128     bf16      256        score@v       34.36       0.56        127.43          56.89    memory\n",
      "128    128     bf16      256        mlp_up(w1)    3848.29     1.23        287.0           2903.49  compute\n",
      "128    128     bf16      256        mlp_down(w2)  3848.29     1.23        287.0           2903.49  compute\n",
      "128    128     bf16      256        mlp_gate(w3)  3848.29     1.23        287.0           2903.49  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       256        qkvo_proj     4398.05     1.06        574.0           3855.06  compute\n",
      "128    128     fp8       256        q@k_T         34.36       0.28        254.86          113.78   memory\n",
      "128    128     fp8       256        softmax       85.9        0.5         22.0            160.0    compute\n",
      "128    128     fp8       256        score@v       34.36       0.28        254.86          113.78   memory\n",
      "128    128     fp8       256        mlp_up(w1)    3848.29     0.62        574.0           5806.99  compute\n",
      "128    128     fp8       256        mlp_down(w2)  3848.29     0.62        574.0           5806.99  compute\n",
      "128    128     fp8       256        mlp_gate(w3)  3848.29     0.62        574.0           5806.99  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     bf16      512        qkvo_proj     8796.09     4.12        287.0           1985.94  compute\n",
      "128    128     bf16      512        q@k_T         68.72       1.12        127.43          56.89    memory\n",
      "128    128     bf16      512        softmax       171.8       1.0         22.0            160.0    compute\n",
      "128    128     bf16      512        score@v       68.72       1.12        127.43          56.89    memory\n",
      "128    128     bf16      512        mlp_up(w1)    7696.58     2.36        287.0           3038.09  compute\n",
      "128    128     bf16      512        mlp_down(w2)  7696.58     2.36        287.0           3038.09  compute\n",
      "128    128     bf16      512        mlp_gate(w3)  7696.58     2.36        287.0           3038.09  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                              prefill                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI       Bound\n",
      "128    128     fp8       512        qkvo_proj     8796.09     2.06        574.0           3971.88  compute\n",
      "128    128     fp8       512        q@k_T         68.72       0.56        254.86          113.78   memory\n",
      "128    128     fp8       512        softmax       171.8       1.0         22.0            160.0    compute\n",
      "128    128     fp8       512        score@v       68.72       0.56        254.86          113.78   memory\n",
      "128    128     fp8       512        mlp_up(w1)    7696.58     1.18        574.0           6076.19  compute\n",
      "128    128     fp8       512        mlp_down(w2)  7696.58     1.18        574.0           6076.19  compute\n",
      "128    128     fp8       512        mlp_gate(w3)  7696.58     1.18        574.0           6076.19  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      1          qkvo_proj     0.13        0.13        2.24            1.0    memory\n",
      "128    128     bf16      1          q@k_T         0.0         0.0         2.24            3.76   compute\n",
      "128    128     bf16      1          softmax       0.0         0.0         22.0            160.0  compute\n",
      "128    128     bf16      1          score@v       0.0         0.0         2.24            3.76   compute\n",
      "128    128     bf16      1          mlp_up(w1)    0.12        0.11        2.24            1.0    memory\n",
      "128    128     bf16      1          mlp_down(w2)  0.12        0.11        2.24            1.0    memory\n",
      "128    128     bf16      1          mlp_gate(w3)  0.12        0.11        2.24            1.0    memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     fp8       1          qkvo_proj     0.13        0.06        4.48            2.0    memory\n",
      "128    128     fp8       1          q@k_T         0.0         0.0         4.48            7.53   compute\n",
      "128    128     fp8       1          softmax       0.0         0.0         22.0            160.0  compute\n",
      "128    128     fp8       1          score@v       0.0         0.0         4.48            7.53   compute\n",
      "128    128     fp8       1          mlp_up(w1)    0.12        0.05        4.48            2.0    memory\n",
      "128    128     fp8       1          mlp_down(w2)  0.12        0.05        4.48            2.0    memory\n",
      "128    128     fp8       1          mlp_gate(w3)  0.12        0.05        4.48            2.0    memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      2          qkvo_proj     0.27        0.13        4.48            2.0    memory\n",
      "128    128     bf16      2          q@k_T         0.0         0.0         4.48            3.76   compute\n",
      "128    128     bf16      2          softmax       0.01        0.0         22.0            160.0  compute\n",
      "128    128     bf16      2          score@v       0.0         0.0         4.48            3.76   compute\n",
      "128    128     bf16      2          mlp_up(w1)    0.23        0.11        4.48            2.0    memory\n",
      "128    128     bf16      2          mlp_down(w2)  0.23        0.11        4.48            2.0    memory\n",
      "128    128     bf16      2          mlp_gate(w3)  0.23        0.11        4.48            2.0    memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     fp8       2          qkvo_proj     0.27        0.06        8.95            4.0    memory\n",
      "128    128     fp8       2          q@k_T         0.0         0.0         8.97            7.53   compute\n",
      "128    128     fp8       2          softmax       0.01        0.0         22.0            160.0  compute\n",
      "128    128     fp8       2          score@v       0.0         0.0         8.97            7.53   compute\n",
      "128    128     fp8       2          mlp_up(w1)    0.23        0.05        8.95            4.0    memory\n",
      "128    128     fp8       2          mlp_down(w2)  0.23        0.05        8.95            4.0    memory\n",
      "128    128     fp8       2          mlp_gate(w3)  0.23        0.05        8.95            4.0    memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      4          qkvo_proj     0.54        0.13        8.94            3.99   memory\n",
      "128    128     bf16      4          q@k_T         0.0         0.0         8.43            3.76   memory\n",
      "128    128     bf16      4          softmax       0.01        0.0         22.0            160.0  compute\n",
      "128    128     bf16      4          score@v       0.0         0.0         8.43            3.76   memory\n",
      "128    128     bf16      4          mlp_up(w1)    0.47        0.11        8.95            3.99   memory\n",
      "128    128     bf16      4          mlp_down(w2)  0.47        0.11        8.95            3.99   memory\n",
      "128    128     bf16      4          mlp_gate(w3)  0.47        0.11        8.95            3.99   memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     fp8       4          qkvo_proj     0.54        0.06        17.89           7.98   memory\n",
      "128    128     fp8       4          q@k_T         0.0         0.0         16.87           7.53   memory\n",
      "128    128     fp8       4          softmax       0.01        0.0         22.0            160.0  compute\n",
      "128    128     fp8       4          score@v       0.0         0.0         16.87           7.53   memory\n",
      "128    128     fp8       4          mlp_up(w1)    0.47        0.05        17.9            7.99   memory\n",
      "128    128     fp8       4          mlp_down(w2)  0.47        0.05        17.9            7.99   memory\n",
      "128    128     fp8       4          mlp_gate(w3)  0.47        0.05        17.9            7.99   memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      8          qkvo_proj     1.07        0.13        17.85           7.97   memory\n",
      "128    128     bf16      8          q@k_T         0.01        0.0         8.43            3.76   memory\n",
      "128    128     bf16      8          softmax       0.02        0.0         22.0            160.0  compute\n",
      "128    128     bf16      8          score@v       0.01        0.0         8.43            3.76   memory\n",
      "128    128     bf16      8          mlp_up(w1)    0.94        0.11        17.88           7.98   memory\n",
      "128    128     bf16      8          mlp_down(w2)  0.94        0.11        17.88           7.98   memory\n",
      "128    128     bf16      8          mlp_gate(w3)  0.94        0.11        17.88           7.98   memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     fp8       8          qkvo_proj     1.07        0.06        35.7            15.94  memory\n",
      "128    128     fp8       8          q@k_T         0.01        0.0         16.87           7.53   memory\n",
      "128    128     fp8       8          softmax       0.02        0.0         22.0            160.0  compute\n",
      "128    128     fp8       8          score@v       0.01        0.0         16.87           7.53   memory\n",
      "128    128     fp8       8          mlp_up(w1)    0.94        0.05        35.75           15.96  memory\n",
      "128    128     fp8       8          mlp_down(w2)  0.94        0.05        35.75           15.96  memory\n",
      "128    128     fp8       8          mlp_gate(w3)  0.94        0.05        35.75           15.96  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      16         qkvo_proj     2.15        0.13        35.56           15.88  memory\n",
      "128    128     bf16      16         q@k_T         0.02        0.0         8.43            3.76   memory\n",
      "128    128     bf16      16         softmax       0.04        0.0         22.0            160.0  compute\n",
      "128    128     bf16      16         score@v       0.02        0.0         8.43            3.76   memory\n",
      "128    128     bf16      16         mlp_up(w1)    1.88        0.11        35.66           15.92  memory\n",
      "128    128     bf16      16         mlp_down(w2)  1.88        0.11        35.66           15.92  memory\n",
      "128    128     bf16      16         mlp_gate(w3)  1.88        0.11        35.66           15.92  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     fp8       16         qkvo_proj     2.15        0.06        71.12           31.75  memory\n",
      "128    128     fp8       16         q@k_T         0.02        0.0         16.87           7.53   memory\n",
      "128    128     fp8       16         softmax       0.04        0.0         22.0            160.0  compute\n",
      "128    128     fp8       16         score@v       0.02        0.0         16.87           7.53   memory\n",
      "128    128     fp8       16         mlp_up(w1)    1.88        0.05        71.32           31.84  memory\n",
      "128    128     fp8       16         mlp_down(w2)  1.88        0.05        71.32           31.84  memory\n",
      "128    128     fp8       16         mlp_gate(w3)  1.88        0.05        71.32           31.84  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      32         qkvo_proj     4.29        0.13        70.58           31.51  memory\n",
      "128    128     bf16      32         q@k_T         0.03        0.01        8.43            3.76   memory\n",
      "128    128     bf16      32         softmax       0.08        0.0         22.0            160.0  compute\n",
      "128    128     bf16      32         score@v       0.03        0.01        8.43            3.76   memory\n",
      "128    128     bf16      32         mlp_up(w1)    3.76        0.11        70.97           31.68  memory\n",
      "128    128     bf16      32         mlp_down(w2)  3.76        0.11        70.97           31.68  memory\n",
      "128    128     bf16      32         mlp_gate(w3)  3.76        0.11        70.97           31.68  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     fp8       32         qkvo_proj     4.29        0.06        141.15          63.02  memory\n",
      "128    128     fp8       32         q@k_T         0.03        0.0         16.87           7.53   memory\n",
      "128    128     fp8       32         softmax       0.08        0.0         22.0            160.0  compute\n",
      "128    128     fp8       32         score@v       0.03        0.0         16.87           7.53   memory\n",
      "128    128     fp8       32         mlp_up(w1)    3.76        0.06        141.93          63.36  memory\n",
      "128    128     fp8       32         mlp_down(w2)  3.76        0.06        141.93          63.36  memory\n",
      "128    128     fp8       32         mlp_gate(w3)  3.76        0.06        141.93          63.36  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI     Bound\n",
      "128    128     bf16      64         qkvo_proj     8.59        0.13        139.02          62.06  memory\n",
      "128    128     bf16      64         q@k_T         0.07        0.02        8.43            3.76   memory\n",
      "128    128     bf16      64         softmax       0.17        0.0         22.0            160.0  compute\n",
      "128    128     bf16      64         score@v       0.07        0.02        8.43            3.76   memory\n",
      "128    128     bf16      64         mlp_up(w1)    7.52        0.11        140.54          62.74  memory\n",
      "128    128     bf16      64         mlp_down(w2)  7.52        0.11        140.54          62.74  memory\n",
      "128    128     bf16      64         mlp_gate(w3)  7.52        0.11        140.54          62.74  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  -----  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       64         qkvo_proj     8.59        0.06        278.03          124.12  memory\n",
      "128    128     fp8       64         q@k_T         0.07        0.01        16.87           7.53    memory\n",
      "128    128     fp8       64         softmax       0.17        0.0         22.0            160.0   compute\n",
      "128    128     fp8       64         score@v       0.07        0.01        16.87           7.53    memory\n",
      "128    128     fp8       64         mlp_up(w1)    7.52        0.06        281.07          125.48  memory\n",
      "128    128     fp8       64         mlp_down(w2)  7.52        0.06        281.07          125.48  memory\n",
      "128    128     fp8       64         mlp_gate(w3)  7.52        0.06        281.07          125.48  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      128        qkvo_proj     17.18       0.13        269.85          120.47  memory\n",
      "128    128     bf16      128        q@k_T         0.13        0.03        8.43            3.76    memory\n",
      "128    128     bf16      128        softmax       0.34        0.0         22.0            160.0   compute\n",
      "128    128     bf16      128        score@v       0.13        0.03        8.43            3.76    memory\n",
      "128    128     bf16      128        mlp_up(w1)    15.03       0.11        275.64          123.06  memory\n",
      "128    128     bf16      128        mlp_down(w2)  15.03       0.11        275.64          123.06  memory\n",
      "128    128     bf16      128        mlp_gate(w3)  15.03       0.11        275.64          123.06  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       128        qkvo_proj     17.18       0.07        539.71          240.94  memory\n",
      "128    128     fp8       128        q@k_T         0.13        0.02        16.87           7.53    memory\n",
      "128    128     fp8       128        softmax       0.34        0.0         22.0            160.0   compute\n",
      "128    128     fp8       128        score@v       0.13        0.02        16.87           7.53    memory\n",
      "128    128     fp8       128        mlp_up(w1)    15.03       0.06        551.29          246.11  memory\n",
      "128    128     fp8       128        mlp_down(w2)  15.03       0.06        551.29          246.11  memory\n",
      "128    128     fp8       128        mlp_gate(w3)  15.03       0.06        551.29          246.11  memory\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      256        qkvo_proj     34.36       0.14        287.0           227.56  compute\n",
      "128    128     bf16      256        q@k_T         0.27        0.07        8.43            3.76    memory\n",
      "128    128     bf16      256        softmax       0.67        0.0         22.0            160.0   compute\n",
      "128    128     bf16      256        score@v       0.27        0.07        8.43            3.76    memory\n",
      "128    128     bf16      256        mlp_up(w1)    30.06       0.12        287.0           236.96  compute\n",
      "128    128     bf16      256        mlp_down(w2)  30.06       0.12        287.0           236.96  compute\n",
      "128    128     bf16      256        mlp_gate(w3)  30.06       0.12        287.0           236.96  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       256        qkvo_proj     34.36       0.07        574.0           455.11  compute\n",
      "128    128     fp8       256        q@k_T         0.27        0.03        16.87           7.53    memory\n",
      "128    128     fp8       256        softmax       0.67        0.0         22.0            160.0   compute\n",
      "128    128     fp8       256        score@v       0.27        0.03        16.87           7.53    memory\n",
      "128    128     fp8       256        mlp_up(w1)    30.06       0.06        574.0           473.92  compute\n",
      "128    128     fp8       256        mlp_down(w2)  30.06       0.06        574.0           473.92  compute\n",
      "128    128     fp8       256        mlp_gate(w3)  30.06       0.06        574.0           473.92  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     bf16      512        qkvo_proj     68.72       0.16        287.0           409.6   compute\n",
      "128    128     bf16      512        q@k_T         0.54        0.13        8.43            3.76    memory\n",
      "128    128     bf16      512        softmax       1.34        0.01        22.0            160.0   compute\n",
      "128    128     bf16      512        score@v       0.54        0.13        8.43            3.76    memory\n",
      "128    128     bf16      512        mlp_up(w1)    60.13       0.13        287.0           441.11  compute\n",
      "128    128     bf16      512        mlp_down(w2)  60.13       0.13        287.0           441.11  compute\n",
      "128    128     bf16      512        mlp_gate(w3)  60.13       0.13        287.0           441.11  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "                                               decode                                               \n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n",
      "Input  Output  DataType  BatchSize  LayerName     NumOps(e9)  Memory(GB)  TopsRF(TFlops)  AI      Bound\n",
      "128    128     fp8       512        qkvo_proj     68.72       0.08        574.0           819.2   compute\n",
      "128    128     fp8       512        q@k_T         0.54        0.07        16.87           7.53    memory\n",
      "128    128     fp8       512        softmax       1.34        0.01        22.0            160.0   compute\n",
      "128    128     fp8       512        score@v       0.54        0.07        16.87           7.53    memory\n",
      "128    128     fp8       512        mlp_up(w1)    60.13       0.06        574.0           882.22  compute\n",
      "128    128     fp8       512        mlp_down(w2)  60.13       0.06        574.0           882.22  compute\n",
      "128    128     fp8       512        mlp_gate(w3)  60.13       0.06        574.0           882.22  compute\n",
      "-----  ------  --------  ---------  ------------  ----------  ----------  --------------  ------  -------\n"
     ]
    }
   ],
   "source": [
    "print_analysis(analysis_dict, batchsize_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "32a2ff1c-181b-450a-8f3b-817713e786f5",
   "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}