DONUT-DocVQA.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "ef83552d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Could not find image processor class in the image processor config or the model config. Loading based on pattern matching with the model's feature extractor configuration.\n"
     ]
    }
   ],
   "source": [
    "from transformers import DonutProcessor, VisionEncoderDecoderModel\n",
    "\n",
    "model_name = \"naver-clova-ix/donut-base-finetuned-docvqa\"\n",
    "processor = DonutProcessor.from_pretrained(model_name)\n",
    "model = VisionEncoderDecoderModel.from_pretrained(model_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "bb268ae7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "VisionEncoderDecoderModel(\n",
       "  (encoder): DonutSwinModel(\n",
       "    (embeddings): DonutSwinEmbeddings(\n",
       "      (patch_embeddings): DonutSwinPatchEmbeddings(\n",
       "        (projection): Conv2d(3, 128, kernel_size=(4, 4), stride=(4, 4))\n",
       "      )\n",
       "      (norm): LayerNorm((128,), eps=1e-05, elementwise_affine=True)\n",
       "      (dropout): Dropout(p=0.0, inplace=False)\n",
       "    )\n",
       "    (encoder): DonutSwinEncoder(\n",
       "      (layers): ModuleList(\n",
       "        (0): DonutSwinStage(\n",
       "          (blocks): ModuleList(\n",
       "            (0-1): 2 x DonutSwinLayer(\n",
       "              (layernorm_before): LayerNorm((128,), eps=1e-05, elementwise_affine=True)\n",
       "              (attention): DonutSwinAttention(\n",
       "                (self): DonutSwinSelfAttention(\n",
       "                  (query): Linear(in_features=128, out_features=128, bias=True)\n",
       "                  (key): Linear(in_features=128, out_features=128, bias=True)\n",
       "                  (value): Linear(in_features=128, out_features=128, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "                (output): DonutSwinSelfOutput(\n",
       "                  (dense): Linear(in_features=128, out_features=128, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "              )\n",
       "              (drop_path): DonutSwinDropPath(p=0.1)\n",
       "              (layernorm_after): LayerNorm((128,), eps=1e-05, elementwise_affine=True)\n",
       "              (intermediate): DonutSwinIntermediate(\n",
       "                (dense): Linear(in_features=128, out_features=512, bias=True)\n",
       "                (intermediate_act_fn): GELUActivation()\n",
       "              )\n",
       "              (output): DonutSwinOutput(\n",
       "                (dense): Linear(in_features=512, out_features=128, bias=True)\n",
       "                (dropout): Dropout(p=0.0, inplace=False)\n",
       "              )\n",
       "            )\n",
       "          )\n",
       "          (downsample): DonutSwinPatchMerging(\n",
       "            (reduction): Linear(in_features=512, out_features=256, bias=False)\n",
       "            (norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)\n",
       "          )\n",
       "        )\n",
       "        (1): DonutSwinStage(\n",
       "          (blocks): ModuleList(\n",
       "            (0-1): 2 x DonutSwinLayer(\n",
       "              (layernorm_before): LayerNorm((256,), eps=1e-05, elementwise_affine=True)\n",
       "              (attention): DonutSwinAttention(\n",
       "                (self): DonutSwinSelfAttention(\n",
       "                  (query): Linear(in_features=256, out_features=256, bias=True)\n",
       "                  (key): Linear(in_features=256, out_features=256, bias=True)\n",
       "                  (value): Linear(in_features=256, out_features=256, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "                (output): DonutSwinSelfOutput(\n",
       "                  (dense): Linear(in_features=256, out_features=256, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "              )\n",
       "              (drop_path): DonutSwinDropPath(p=0.1)\n",
       "              (layernorm_after): LayerNorm((256,), eps=1e-05, elementwise_affine=True)\n",
       "              (intermediate): DonutSwinIntermediate(\n",
       "                (dense): Linear(in_features=256, out_features=1024, bias=True)\n",
       "                (intermediate_act_fn): GELUActivation()\n",
       "              )\n",
       "              (output): DonutSwinOutput(\n",
       "                (dense): Linear(in_features=1024, out_features=256, bias=True)\n",
       "                (dropout): Dropout(p=0.0, inplace=False)\n",
       "              )\n",
       "            )\n",
       "          )\n",
       "          (downsample): DonutSwinPatchMerging(\n",
       "            (reduction): Linear(in_features=1024, out_features=512, bias=False)\n",
       "            (norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "          )\n",
       "        )\n",
       "        (2): DonutSwinStage(\n",
       "          (blocks): ModuleList(\n",
       "            (0-13): 14 x DonutSwinLayer(\n",
       "              (layernorm_before): LayerNorm((512,), eps=1e-05, elementwise_affine=True)\n",
       "              (attention): DonutSwinAttention(\n",
       "                (self): DonutSwinSelfAttention(\n",
       "                  (query): Linear(in_features=512, out_features=512, bias=True)\n",
       "                  (key): Linear(in_features=512, out_features=512, bias=True)\n",
       "                  (value): Linear(in_features=512, out_features=512, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "                (output): DonutSwinSelfOutput(\n",
       "                  (dense): Linear(in_features=512, out_features=512, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "              )\n",
       "              (drop_path): DonutSwinDropPath(p=0.1)\n",
       "              (layernorm_after): LayerNorm((512,), eps=1e-05, elementwise_affine=True)\n",
       "              (intermediate): DonutSwinIntermediate(\n",
       "                (dense): Linear(in_features=512, out_features=2048, bias=True)\n",
       "                (intermediate_act_fn): GELUActivation()\n",
       "              )\n",
       "              (output): DonutSwinOutput(\n",
       "                (dense): Linear(in_features=2048, out_features=512, bias=True)\n",
       "                (dropout): Dropout(p=0.0, inplace=False)\n",
       "              )\n",
       "            )\n",
       "          )\n",
       "          (downsample): DonutSwinPatchMerging(\n",
       "            (reduction): Linear(in_features=2048, out_features=1024, bias=False)\n",
       "            (norm): LayerNorm((2048,), eps=1e-05, elementwise_affine=True)\n",
       "          )\n",
       "        )\n",
       "        (3): DonutSwinStage(\n",
       "          (blocks): ModuleList(\n",
       "            (0-1): 2 x DonutSwinLayer(\n",
       "              (layernorm_before): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "              (attention): DonutSwinAttention(\n",
       "                (self): DonutSwinSelfAttention(\n",
       "                  (query): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "                  (key): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "                  (value): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "                (output): DonutSwinSelfOutput(\n",
       "                  (dense): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "                  (dropout): Dropout(p=0.0, inplace=False)\n",
       "                )\n",
       "              )\n",
       "              (drop_path): DonutSwinDropPath(p=0.1)\n",
       "              (layernorm_after): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "              (intermediate): DonutSwinIntermediate(\n",
       "                (dense): Linear(in_features=1024, out_features=4096, bias=True)\n",
       "                (intermediate_act_fn): GELUActivation()\n",
       "              )\n",
       "              (output): DonutSwinOutput(\n",
       "                (dense): Linear(in_features=4096, out_features=1024, bias=True)\n",
       "                (dropout): Dropout(p=0.0, inplace=False)\n",
       "              )\n",
       "            )\n",
       "          )\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "    (pooler): AdaptiveAvgPool1d(output_size=1)\n",
       "  )\n",
       "  (decoder): MBartForCausalLM(\n",
       "    (model): MBartDecoderWrapper(\n",
       "      (decoder): MBartDecoder(\n",
       "        (embed_tokens): Embedding(57532, 1024, padding_idx=1)\n",
       "        (embed_positions): MBartLearnedPositionalEmbedding(130, 1024)\n",
       "        (layers): ModuleList(\n",
       "          (0-3): 4 x MBartDecoderLayer(\n",
       "            (self_attn): MBartAttention(\n",
       "              (k_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "              (v_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "              (q_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "              (out_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "            )\n",
       "            (activation_fn): GELUActivation()\n",
       "            (self_attn_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "            (encoder_attn): MBartAttention(\n",
       "              (k_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "              (v_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "              (q_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "              (out_proj): Linear(in_features=1024, out_features=1024, bias=True)\n",
       "            )\n",
       "            (encoder_attn_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "            (fc1): Linear(in_features=1024, out_features=4096, bias=True)\n",
       "            (fc2): Linear(in_features=4096, out_features=1024, bias=True)\n",
       "            (final_layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "          )\n",
       "        )\n",
       "        (layernorm_embedding): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "        (layer_norm): LayerNorm((1024,), eps=1e-05, elementwise_affine=True)\n",
       "      )\n",
       "    )\n",
       "    (lm_head): Linear(in_features=1024, out_features=57532, bias=False)\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.to(\"cpu\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "f2453056",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=480x524>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from PIL import Image\n",
    "\n",
    "image = Image.open(\"time2.png\")\n",
    "image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "3907ff56",
   "metadata": {},
   "outputs": [],
   "source": [
    "import cv2\n",
    "\n",
    "image = cv2.imread(\"time2.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "177005cd",
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = processor(image, return_tensors=\"pt\").pixel_values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "80c4b479",
   "metadata": {},
   "outputs": [],
   "source": [
    "question = \"When is the Mathematics-I?\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "cf46bce2",
   "metadata": {},
   "outputs": [],
   "source": [
    "prompt = f\"\"\"<s_docvqa><s_question>{question}</s_question><s_answer>\"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "e609f5a6",
   "metadata": {},
   "outputs": [],
   "source": [
    "decoder_input_ids = processor.tokenizer(prompt, add_special_tokens=False, return_tensors=\"pt\")[\"input_ids\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "26f55113",
   "metadata": {},
   "outputs": [],
   "source": [
    "outputs = model.generate(\n",
    "    inputs.to(\"cpu\"),\n",
    "    decoder_input_ids=decoder_input_ids.to(\"cpu\"),\n",
    "    max_length=model.decoder.config.max_position_embeddings,\n",
    "    output_scores=True,\n",
    "    early_stopping=True,\n",
    "    return_dict_in_generate=True,\n",
    "    num_beams=1,\n",
    "    pad_token_id=processor.tokenizer.pad_token_id,\n",
    "    eos_token_id=processor.tokenizer.eos_token_id,\n",
    "    bad_words_ids=[[processor.tokenizer.unk_token_id]],\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "fc9c8bd9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<s_question> When is the Mathematics-I?</s_question><s_answer> 03:30 pm to 05:30 pm</s_answer>\n"
     ]
    }
   ],
   "source": [
    "import re\n",
    "\n",
    "seq = processor.batch_decode(outputs.sequences)[0]\n",
    "seq = seq.replace(processor.tokenizer.eos_token, \"\")\n",
    "seq = seq.replace(processor.tokenizer.pad_token, \"\")\n",
    "seq = re.sub(r\"<.*?>\", \"\", seq, count=1).strip()\n",
    "print(seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "87f1c618",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'question': 'When is the Mathematics-I?', 'answer': '03:30 pm to 05:30 pm'}"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "processor.token2json(seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c64114a6",
   "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": "3.11.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}