{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "7c02ae76-b540-4b16-9492-e9210b3b9249", "metadata": {}, "outputs": [], "source": [ "import os\n", "os.environ['CUDA_VISIBLE_DEVICE'] = ''\n", "import random\n", "\n", "%matplotlib inline\n", "import matplotlib.pyplot as plt\n", "\n", "import torch\n", "import numpy as np\n", "from omegaconf import OmegaConf\n", "\n", "%load_ext autoreload\n", "%autoreload 2\n", "\n", "from importlib.util import find_spec\n", "if find_spec(\"text_recognizer\") is None:\n", " import sys\n", " sys.path.append('..')" ] }, { "cell_type": "code", "execution_count": 2, "id": "ccdb6dde-47e5-429a-88f2-0764fb7e259a", "metadata": {}, "outputs": [], "source": [ "from hydra import compose, initialize\n", "from omegaconf import OmegaConf\n", "from hydra.utils import instantiate" ] }, { "cell_type": "code", "execution_count": 3, "id": "3cf50475-39f2-4642-a7d1-5bcbc0a036f7", "metadata": {}, "outputs": [], "source": [ "path = \"../training/conf/experiment/conv_transformer_paragraphs.yaml\"" ] }, { "cell_type": "code", "execution_count": 86, "id": "e52ecb01-c975-4e55-925d-1182c7aea473", "metadata": {}, "outputs": [], "source": [ "with open(path, \"rb\") as f:\n", " cfg = OmegaConf.load(f)" ] }, { "cell_type": "code", "execution_count": 87, "id": "f939aa37-7b1d-45cc-885c-323c4540bda1", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'dim': 192, 'num_heads': 4, 'dim_head': 64, 'dropout_rate': 0.05, '_target_': 'text_recognizer.networks.transformer.attention.Attention', 'causal': False}" ] }, "execution_count": 87, "metadata": {}, "output_type": "execute_result" } ], "source": [ "cfg.network.decoder.cross_attn" ] }, { "cell_type": "code", "execution_count": 88, "id": "aaeab329-aeb0-4a1b-aa35-5a2aab81b1d0", "metadata": {}, "outputs": [], "source": [ "net = instantiate(cfg.network)" ] }, { "cell_type": "code", "execution_count": 89, "id": "618b997c-e6a6-4487-b70c-9d260cb556d3", "metadata": {}, "outputs": [], "source": [ "from torchinfo import summary" ] }, { "cell_type": "code", "execution_count": 91, "id": "66118c10-2e59-469f-99d6-ddea4bfd0d73", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "====================================================================================================\n", "Layer (type:depth-idx) Output Shape Param #\n", "====================================================================================================\n", "ConvTransformer -- --\n", "├─EfficientNet: 1 -- --\n", "│ └─ModuleList: 2-1 -- --\n", "├─Decoder: 1 -- --\n", "│ └─ModuleList: 2-2 -- --\n", "│ │ └─ModuleList: 3-1 -- --\n", "│ │ └─ModuleList: 3-2 -- --\n", "│ │ └─ModuleList: 3-3 -- --\n", "│ │ └─ModuleList: 3-4 -- --\n", "│ │ └─ModuleList: 3-5 -- --\n", "│ │ └─ModuleList: 3-6 -- --\n", "│ │ └─ModuleList: 3-7 -- --\n", "│ │ └─ModuleList: 3-8 -- --\n", "│ │ └─ModuleList: 3-9 -- --\n", "├─EfficientNet: 1-1 [2, 1280, 18, 20] --\n", "│ └─Sequential: 2-3 [2, 32, 288, 320] --\n", "│ │ └─ZeroPad2d: 3-10 [2, 1, 577, 641] --\n", "│ │ └─Conv2d: 3-11 [2, 32, 288, 320] 288\n", "│ │ └─BatchNorm2d: 3-12 [2, 32, 288, 320] 64\n", "│ │ └─Mish: 3-13 [2, 32, 288, 320] --\n", "│ └─ModuleList: 2-1 -- --\n", "│ │ └─MBConvBlock: 3-14 [2, 16, 288, 320] --\n", "│ │ │ └─Sequential: 4-1 [2, 32, 288, 320] 352\n", "│ │ │ └─Sequential: 4-2 [2, 32, 288, 320] 552\n", "│ │ │ └─Sequential: 4-3 [2, 16, 288, 320] 544\n", "│ │ └─MBConvBlock: 3-15 [2, 24, 144, 160] --\n", "│ │ │ └─Sequential: 4-4 [2, 96, 288, 320] 1,728\n", "│ │ │ └─Sequential: 4-5 [2, 96, 144, 160] 1,056\n", "│ │ │ └─Sequential: 4-6 [2, 96, 144, 160] 4,728\n", "│ │ │ └─Sequential: 4-7 [2, 24, 144, 160] 2,352\n", "│ │ └─MBConvBlock: 3-16 [2, 24, 144, 160] --\n", "│ │ │ └─Sequential: 4-8 [2, 144, 144, 160] 3,744\n", "│ │ │ └─Sequential: 4-9 [2, 144, 144, 160] 1,584\n", "│ │ │ └─Sequential: 4-10 [2, 144, 144, 160] 10,548\n", "│ │ │ └─Sequential: 4-11 [2, 24, 144, 160] 3,504\n", "│ │ └─MBConvBlock: 3-17 [2, 40, 72, 80] --\n", "│ │ │ └─Sequential: 4-12 [2, 144, 144, 160] 3,744\n", "│ │ │ └─Sequential: 4-13 [2, 144, 72, 80] 3,888\n", "│ │ │ └─Sequential: 4-14 [2, 144, 72, 80] 10,548\n", "│ │ │ └─Sequential: 4-15 [2, 40, 72, 80] 5,840\n", "│ │ └─MBConvBlock: 3-18 [2, 40, 72, 80] --\n", "│ │ │ └─Sequential: 4-16 [2, 240, 72, 80] 10,080\n", "│ │ │ └─Sequential: 4-17 [2, 240, 72, 80] 6,480\n", "│ │ │ └─Sequential: 4-18 [2, 240, 72, 80] 29,100\n", "│ │ │ └─Sequential: 4-19 [2, 40, 72, 80] 9,680\n", "│ │ └─MBConvBlock: 3-19 [2, 80, 36, 40] --\n", "│ │ │ └─Sequential: 4-20 [2, 240, 72, 80] 10,080\n", "│ │ │ └─Sequential: 4-21 [2, 240, 36, 40] 2,640\n", "│ │ │ └─Sequential: 4-22 [2, 240, 36, 40] 29,100\n", "│ │ │ └─Sequential: 4-23 [2, 80, 36, 40] 19,360\n", "│ │ └─MBConvBlock: 3-20 [2, 80, 36, 40] --\n", "│ │ │ └─Sequential: 4-24 [2, 480, 36, 40] 39,360\n", "│ │ │ └─Sequential: 4-25 [2, 480, 36, 40] 5,280\n", "│ │ │ └─Sequential: 4-26 [2, 480, 36, 40] 115,800\n", "│ │ │ └─Sequential: 4-27 [2, 80, 36, 40] 38,560\n", "│ │ └─MBConvBlock: 3-21 [2, 80, 36, 40] --\n", "│ │ │ └─Sequential: 4-28 [2, 480, 36, 40] 39,360\n", "│ │ │ └─Sequential: 4-29 [2, 480, 36, 40] 5,280\n", "│ │ │ └─Sequential: 4-30 [2, 480, 36, 40] 115,800\n", "│ │ │ └─Sequential: 4-31 [2, 80, 36, 40] 38,560\n", "│ │ └─MBConvBlock: 3-22 [2, 112, 36, 40] --\n", "│ │ │ └─Sequential: 4-32 [2, 480, 36, 40] 39,360\n", "│ │ │ └─Sequential: 4-33 [2, 480, 36, 40] 12,960\n", "│ │ │ └─Sequential: 4-34 [2, 480, 36, 40] 115,800\n", "│ │ │ └─Sequential: 4-35 [2, 112, 36, 40] 53,984\n", "│ │ └─MBConvBlock: 3-23 [2, 112, 36, 40] --\n", "│ │ │ └─Sequential: 4-36 [2, 672, 36, 40] 76,608\n", "│ │ │ └─Sequential: 4-37 [2, 672, 36, 40] 18,144\n", "│ │ │ └─Sequential: 4-38 [2, 672, 36, 40] 226,632\n", "│ │ │ └─Sequential: 4-39 [2, 112, 36, 40] 75,488\n", "│ │ └─MBConvBlock: 3-24 [2, 112, 36, 40] --\n", "│ │ │ └─Sequential: 4-40 [2, 672, 36, 40] 76,608\n", "│ │ │ └─Sequential: 4-41 [2, 672, 36, 40] 18,144\n", "│ │ │ └─Sequential: 4-42 [2, 672, 36, 40] 226,632\n", "│ │ │ └─Sequential: 4-43 [2, 112, 36, 40] 75,488\n", "│ │ └─MBConvBlock: 3-25 [2, 192, 18, 20] --\n", "│ │ │ └─Sequential: 4-44 [2, 672, 36, 40] 76,608\n", "│ │ │ └─Sequential: 4-45 [2, 672, 18, 20] 18,144\n", "│ │ │ └─Sequential: 4-46 [2, 672, 18, 20] 226,632\n", "│ │ │ └─Sequential: 4-47 [2, 192, 18, 20] 129,408\n", "│ │ └─MBConvBlock: 3-26 [2, 192, 18, 20] --\n", "│ │ │ └─Sequential: 4-48 [2, 1152, 18, 20] 223,488\n", "│ │ │ └─Sequential: 4-49 [2, 1152, 18, 20] 31,104\n", "│ │ │ └─Sequential: 4-50 [2, 1152, 18, 20] 664,992\n", "│ │ │ └─Sequential: 4-51 [2, 192, 18, 20] 221,568\n", "│ │ └─MBConvBlock: 3-27 [2, 192, 18, 20] --\n", "│ │ │ └─Sequential: 4-52 [2, 1152, 18, 20] 223,488\n", "│ │ │ └─Sequential: 4-53 [2, 1152, 18, 20] 31,104\n", "│ │ │ └─Sequential: 4-54 [2, 1152, 18, 20] 664,992\n", "│ │ │ └─Sequential: 4-55 [2, 192, 18, 20] 221,568\n", "│ │ └─MBConvBlock: 3-28 [2, 192, 18, 20] --\n", "│ │ │ └─Sequential: 4-56 [2, 1152, 18, 20] 223,488\n", "│ │ │ └─Sequential: 4-57 [2, 1152, 18, 20] 31,104\n", "│ │ │ └─Sequential: 4-58 [2, 1152, 18, 20] 664,992\n", "│ │ │ └─Sequential: 4-59 [2, 192, 18, 20] 221,568\n", "│ │ └─MBConvBlock: 3-29 [2, 320, 18, 20] --\n", "│ │ │ └─Sequential: 4-60 [2, 1152, 18, 20] 223,488\n", "│ │ │ └─Sequential: 4-61 [2, 1152, 18, 20] 12,672\n", "│ │ │ └─Sequential: 4-62 [2, 1152, 18, 20] 664,992\n", "│ │ │ └─Sequential: 4-63 [2, 320, 18, 20] 369,280\n", "│ └─Sequential: 2-4 [2, 1280, 18, 20] --\n", "│ │ └─Conv2d: 3-30 [2, 1280, 18, 20] 409,600\n", "│ │ └─BatchNorm2d: 3-31 [2, 1280, 18, 20] 2,560\n", "├─Sequential: 1-2 [2, 192, 360] --\n", "│ └─Conv2d: 2-5 [2, 192, 18, 20] 245,952\n", "│ └─AxialPositionalEmbedding: 2-6 [2, 192, 18, 20] 7,296\n", "│ └─Flatten: 2-7 [2, 192, 360] --\n", "├─Embedding: 1-3 [1, 682, 192] 11,136\n", "├─Decoder: 1-4 [2, 682, 192] --\n", "│ └─ModuleList: 2-2 -- --\n", "│ │ └─ModuleList: 3-1 -- --\n", "│ │ │ └─ScaleNorm: 4-64 [1, 682, 192] 1\n", "│ │ │ └─LocalAttention: 4-65 [1, 682, 192] 196,800\n", "│ │ │ └─Residual: 4-66 [1, 682, 192] --\n", "│ │ └─ModuleList: 3-2 -- --\n", "│ │ │ └─ScaleNorm: 4-67 [1, 682, 192] 1\n", "│ │ │ └─Attention: 4-68 [2, 682, 192] 196,800\n", "│ │ │ └─Residual: 4-69 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-3 -- --\n", "│ │ │ └─ScaleNorm: 4-70 [2, 682, 192] 1\n", "│ │ │ └─FeedForward: 4-71 [2, 682, 192] 444,096\n", "│ │ │ └─Residual: 4-72 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-4 -- --\n", "│ │ │ └─ScaleNorm: 4-73 [2, 682, 192] 1\n", "│ │ │ └─LocalAttention: 4-74 [2, 682, 192] 196,800\n", "│ │ │ └─Residual: 4-75 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-5 -- --\n", "│ │ │ └─ScaleNorm: 4-76 [2, 682, 192] 1\n", "│ │ │ └─Attention: 4-77 [2, 682, 192] 196,800\n", "│ │ │ └─Residual: 4-78 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-6 -- --\n", "│ │ │ └─ScaleNorm: 4-79 [2, 682, 192] 1\n", "│ │ │ └─FeedForward: 4-80 [2, 682, 192] 444,096\n", "│ │ │ └─Residual: 4-81 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-7 -- --\n", "│ │ │ └─ScaleNorm: 4-82 [2, 682, 192] 1\n", "│ │ │ └─Attention: 4-83 [2, 682, 192] 196,800\n", "│ │ │ └─Residual: 4-84 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-8 -- --\n", "│ │ │ └─ScaleNorm: 4-85 [2, 682, 192] 1\n", "│ │ │ └─Attention: 4-86 [2, 682, 192] 196,800\n", "│ │ │ └─Residual: 4-87 [2, 682, 192] --\n", "│ │ └─ModuleList: 3-9 -- --\n", "│ │ │ └─ScaleNorm: 4-88 [2, 682, 192] 1\n", "│ │ │ └─FeedForward: 4-89 [2, 682, 192] 444,096\n", "│ │ │ └─Residual: 4-90 [2, 682, 192] --\n", "├─Linear: 1-5 [2, 682, 58] 11,194\n", "====================================================================================================\n", "Total params: 9,930,947\n", "Trainable params: 9,930,947\n", "Non-trainable params: 0\n", "Total mult-adds (G): 11.45\n", "====================================================================================================\n", "Input size (MB): 2.95\n", "Forward/backward pass size (MB): 2048.49\n", "Params size (MB): 39.72\n", "Estimated Total Size (MB): 2091.17\n", "====================================================================================================" ] }, "execution_count": 91, "metadata": {}, "output_type": "execute_result" } ], "source": [ "summary(net, ((2, 1, 576, 640), (1, 682)), device=\"cpu\", depth=4)" ] }, { "cell_type": "code", "execution_count": 2, "id": "b13ac47c-322d-47d4-bcee-43e5341f74a7", "metadata": {}, "outputs": [], "source": [ "start_tokens = torch.ones(1, 1).long()" ] }, { "cell_type": "code", "execution_count": 4, "id": "55a16f5d-2b27-4a12-b5bb-eb079784b0ae", "metadata": {}, "outputs": [], "source": [ "num_dims = len(start_tokens.shape)" ] }, { "cell_type": "code", "execution_count": 5, "id": "46c65400-fa47-4c10-9edd-8416e6a1185a", "metadata": {}, "outputs": [], "source": [ "if num_dims == 1:\n", " start_tokens = start_tokens[None, :]" ] }, { "cell_type": "code", "execution_count": 7, "id": "1dfa0b95-a075-4121-b2bf-f1a8100b10fd", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "torch.Size([1, 1])" ] }, "execution_count": 7, "metadata": {}, "output_type": "execute_result" } ], "source": [ "start_tokens.shape" ] }, { "cell_type": "code", "execution_count": 8, "id": "c85a357c-f6af-42c5-b714-89df024c29e3", "metadata": {}, "outputs": [], "source": [ "b, t = start_tokens.shape" ] }, { "cell_type": "code", "execution_count": 10, "id": "0ba293f4-e08d-4aaa-94d5-da4899f9b592", "metadata": {}, "outputs": [], "source": [ "out = start_tokens" ] }, { "cell_type": "code", "execution_count": 11, "id": "a8225f98-c5e9-4da2-b756-75599fa8e044", "metadata": {}, "outputs": [], "source": [ "input_mask = torch.full_like(out, True, dtype=torch.bool, device=out.device)" ] }, { "cell_type": "code", "execution_count": 12, "id": "63dfcfd7-6b93-49ac-a0ab-59be53fa0853", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "tensor([[True]])" ] }, "execution_count": 12, "metadata": {}, "output_type": "execute_result" } ], "source": [ "input_mask" ] }, { "cell_type": "code", "execution_count": 13, "id": "a752bfcf-f323-43bc-a910-fec4695150e0", "metadata": {}, "outputs": [], "source": [ "x = out[:, -200:]" ] }, { "cell_type": "code", "execution_count": 14, "id": "4b1b1989-930a-48c5-a7b3-746289107b97", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "tensor([[1]])" ] }, "execution_count": 14, "metadata": {}, "output_type": "execute_result" } ], "source": [ "x" ] }, { "cell_type": "code", "execution_count": null, "id": "513f27bb-2ae1-42a0-8de9-9ae39fdfff32", "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": "3.9.7" } }, "nbformat": 4, "nbformat_minor": 5 }