{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/huggingface/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n",
      "/home/huggingface/lib/python3.10/site-packages/huggingface_hub/file_download.py:1150: FutureWarning: `resume_download` is deprecated and will be removed in version 1.0.0. Downloads always resume when possible. If you want to force a new download, use `force_download=True`.\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from transformers import AutoTokenizer, AutoModelForSequenceClassification\n",
    "\n",
    "checkpoint = \"distilbert-base-uncased-finetuned-sst-2-english\"\n",
    "tokenizer = AutoTokenizer.from_pretrained(checkpoint)\n",
    "model = AutoModelForSequenceClassification.from_pretrained(checkpoint)\n",
    "model.eval();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Try forward pass on single Example"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([2057, 2342, 2062, 3737, 7435, 1010, 6145, 1998, 9559, 1999, 2256, 3842,\n",
       "        1012])"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sequence = \"We need more quality doctors, engineers and lawyers in our nation.\"\n",
    "token_ids = torch.tensor(tokenizer.convert_tokens_to_ids(tokenizer.tokenize(sequence)))\n",
    "token_ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "ename": "RuntimeError",
     "evalue": "The size of tensor a (13) must match the size of tensor b (512) at non-singleton dimension 1",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mRuntimeError\u001b[0m                              Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[3], line 2\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m torch\u001b[38;5;241m.\u001b[39mno_grad():\n\u001b[0;32m----> 2\u001b[0m     \u001b[43mmodel\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtoken_ids\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/torch/nn/modules/module.py:1532\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1530\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m   1531\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1532\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/torch/nn/modules/module.py:1541\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1536\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1537\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1538\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1539\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1540\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1541\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1543\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m   1544\u001b[0m     result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/transformers/models/distilbert/modeling_distilbert.py:763\u001b[0m, in \u001b[0;36mDistilBertForSequenceClassification.forward\u001b[0;34m(self, input_ids, attention_mask, head_mask, inputs_embeds, labels, output_attentions, output_hidden_states, return_dict)\u001b[0m\n\u001b[1;32m    755\u001b[0m \u001b[38;5;250m\u001b[39m\u001b[38;5;124mr\u001b[39m\u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m    756\u001b[0m \u001b[38;5;124;03mlabels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):\u001b[39;00m\n\u001b[1;32m    757\u001b[0m \u001b[38;5;124;03m    Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,\u001b[39;00m\n\u001b[1;32m    758\u001b[0m \u001b[38;5;124;03m    config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If\u001b[39;00m\n\u001b[1;32m    759\u001b[0m \u001b[38;5;124;03m    `config.num_labels > 1` a classification loss is computed (Cross-Entropy).\u001b[39;00m\n\u001b[1;32m    760\u001b[0m \u001b[38;5;124;03m\"\"\"\u001b[39;00m\n\u001b[1;32m    761\u001b[0m return_dict \u001b[38;5;241m=\u001b[39m return_dict \u001b[38;5;28;01mif\u001b[39;00m return_dict \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;28;01melse\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mconfig\u001b[38;5;241m.\u001b[39muse_return_dict\n\u001b[0;32m--> 763\u001b[0m distilbert_output \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdistilbert\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    764\u001b[0m \u001b[43m    \u001b[49m\u001b[43minput_ids\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43minput_ids\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    765\u001b[0m \u001b[43m    \u001b[49m\u001b[43mattention_mask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mattention_mask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    766\u001b[0m \u001b[43m    \u001b[49m\u001b[43mhead_mask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mhead_mask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    767\u001b[0m \u001b[43m    \u001b[49m\u001b[43minputs_embeds\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43minputs_embeds\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    768\u001b[0m \u001b[43m    \u001b[49m\u001b[43moutput_attentions\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moutput_attentions\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    769\u001b[0m \u001b[43m    \u001b[49m\u001b[43moutput_hidden_states\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moutput_hidden_states\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    770\u001b[0m \u001b[43m    \u001b[49m\u001b[43mreturn_dict\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mreturn_dict\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    771\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    772\u001b[0m hidden_state \u001b[38;5;241m=\u001b[39m distilbert_output[\u001b[38;5;241m0\u001b[39m]  \u001b[38;5;66;03m# (bs, seq_len, dim)\u001b[39;00m\n\u001b[1;32m    773\u001b[0m pooled_output \u001b[38;5;241m=\u001b[39m hidden_state[:, \u001b[38;5;241m0\u001b[39m]  \u001b[38;5;66;03m# (bs, dim)\u001b[39;00m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/torch/nn/modules/module.py:1532\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1530\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m   1531\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1532\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/torch/nn/modules/module.py:1541\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1536\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1537\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1538\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1539\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1540\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1541\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1543\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m   1544\u001b[0m     result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/transformers/models/distilbert/modeling_distilbert.py:581\u001b[0m, in \u001b[0;36mDistilBertModel.forward\u001b[0;34m(self, input_ids, attention_mask, head_mask, inputs_embeds, output_attentions, output_hidden_states, return_dict)\u001b[0m\n\u001b[1;32m    578\u001b[0m \u001b[38;5;66;03m# Prepare head mask if needed\u001b[39;00m\n\u001b[1;32m    579\u001b[0m head_mask \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mget_head_mask(head_mask, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mconfig\u001b[38;5;241m.\u001b[39mnum_hidden_layers)\n\u001b[0;32m--> 581\u001b[0m embeddings \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43membeddings\u001b[49m\u001b[43m(\u001b[49m\u001b[43minput_ids\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43minputs_embeds\u001b[49m\u001b[43m)\u001b[49m  \u001b[38;5;66;03m# (bs, seq_length, dim)\u001b[39;00m\n\u001b[1;32m    583\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mtransformer(\n\u001b[1;32m    584\u001b[0m     x\u001b[38;5;241m=\u001b[39membeddings,\n\u001b[1;32m    585\u001b[0m     attn_mask\u001b[38;5;241m=\u001b[39mattention_mask,\n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    589\u001b[0m     return_dict\u001b[38;5;241m=\u001b[39mreturn_dict,\n\u001b[1;32m    590\u001b[0m )\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/torch/nn/modules/module.py:1532\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1530\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m   1531\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1532\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/torch/nn/modules/module.py:1541\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1536\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1537\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1538\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1539\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1540\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1541\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1543\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m   1544\u001b[0m     result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[0;32m~/huggingface/lib/python3.10/site-packages/transformers/models/distilbert/modeling_distilbert.py:135\u001b[0m, in \u001b[0;36mEmbeddings.forward\u001b[0;34m(self, input_ids, input_embeds)\u001b[0m\n\u001b[1;32m    131\u001b[0m     position_ids \u001b[38;5;241m=\u001b[39m position_ids\u001b[38;5;241m.\u001b[39munsqueeze(\u001b[38;5;241m0\u001b[39m)\u001b[38;5;241m.\u001b[39mexpand_as(input_ids)  \u001b[38;5;66;03m# (bs, max_seq_length)\u001b[39;00m\n\u001b[1;32m    133\u001b[0m position_embeddings \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mposition_embeddings(position_ids)  \u001b[38;5;66;03m# (bs, max_seq_length, dim)\u001b[39;00m\n\u001b[0;32m--> 135\u001b[0m embeddings \u001b[38;5;241m=\u001b[39m \u001b[43minput_embeds\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m+\u001b[39;49m\u001b[43m \u001b[49m\u001b[43mposition_embeddings\u001b[49m  \u001b[38;5;66;03m# (bs, max_seq_length, dim)\u001b[39;00m\n\u001b[1;32m    136\u001b[0m embeddings \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mLayerNorm(embeddings)  \u001b[38;5;66;03m# (bs, max_seq_length, dim)\u001b[39;00m\n\u001b[1;32m    137\u001b[0m embeddings \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mdropout(embeddings)  \u001b[38;5;66;03m# (bs, max_seq_length, dim)\u001b[39;00m\n",
      "\u001b[0;31mRuntimeError\u001b[0m: The size of tensor a (13) must match the size of tensor b (512) at non-singleton dimension 1"
     ]
    }
   ],
   "source": [
    "with torch.no_grad():\n",
    "    model(token_ids)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([2057, 2342, 2062, 3737, 7435, 1010, 6145, 1998, 9559, 1999, 2256, 3842,\n",
       "        1012])"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "token_ids"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As seen above our model does not have a batch dimension because of which we are seeing this issue. Let's add a batch dimension and then pass our sequence through the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SequenceClassifierOutput(loss=None, logits=tensor([[ 1.2781, -1.0656]]), hidden_states=None, attentions=None)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "with torch.no_grad():\n",
    "    out = model(token_ids.unsqueeze(0))\n",
    "out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's try by duplicating the input if we get the same logits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 1.2781, -1.0656],\n",
       "        [ 1.2781, -1.0656]])"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "with torch.no_grad():\n",
    "    inp = torch.cat([token_ids.unsqueeze(0), token_ids.unsqueeze(0)], dim = 0)\n",
    "    out = model(inp)\n",
    "out.logits"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Input padding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[ 1.5694, -1.3895]], grad_fn=<AddmmBackward0>)\n",
      "tensor([[ 0.5803, -0.4125]], grad_fn=<AddmmBackward0>)\n",
      "tensor([[ 1.5694, -1.3895],\n",
      "        [ 0.9907, -0.9139]], grad_fn=<AddmmBackward0>)\n"
     ]
    }
   ],
   "source": [
    "padding_id = 100\n",
    "\n",
    "batched_ids = [\n",
    "    [200, 200, 200],\n",
    "    [200, 200, padding_id],\n",
    "]\n",
    "\n",
    "print(model(torch.tensor([batched_ids[0]])).logits)\n",
    "print(model(torch.tensor([batched_ids[1][:2]])).logits)\n",
    "print(model(torch.tensor(batched_ids)).logits)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "There’s something wrong with the logits in our batched predictions: the second row should be the same as the logits for the second sentence, but we’ve got completely different values!\n",
    "\n",
    "This is because when we add padding, we need to make sure we nullify it's impact during the attention matrix computation step. This is why we need a mask so that we can explicily shut these tokens from the attention calculation."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cross checking the working of attention masks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'input_ids': tensor([[  101,  1045,  1521,  2310,  2042,  3403,  2005,  1037, 17662, 12172,\n",
       "          2607,  2026,  2878,  2166,  1012,   102],\n",
       "        [  101,  1045,  5223,  2023,  2061,  2172,   999,   102,     0,     0,\n",
       "             0,     0,     0,     0,     0,     0]]), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n",
       "        [1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]])}"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "sentences = [\"I’ve been waiting for a HuggingFace course my whole life.\",\n",
    "             \"I hate this so much!\"]\n",
    "tokens = tokenizer(sentences, padding=True, return_tensors=\"pt\")\n",
    "with torch.no_grad():\n",
    "    out = model(**tokens)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'input_ids': tensor([[  101,  1045,  1521,  2310,  2042,  3403,  2005,  1037, 17662, 12172,\n",
      "          2607,  2026,  2878,  2166,  1012,   102],\n",
      "        [  101,  1045,  5223,  2023,  2061,  2172,   999,   102,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0]]), 'attention_mask': tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],\n",
      "        [1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0]])}\n"
     ]
    }
   ],
   "source": [
    "print(tokens)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-1.5979,  1.6390],\n",
       "        [ 4.1692, -3.3464]])"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out.logits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[-1.5979,  1.6390]])\n"
     ]
    }
   ],
   "source": [
    "# Do the entire forward pass manually for sentence 1\n",
    "\n",
    "# Tokenize the sentence and get the tokenids\n",
    "token_ids = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(sentences[0]))\n",
    "\n",
    "# Add the special token CLS and SEP at the start and end of the token rspectively\n",
    "token_ids = [101] + token_ids + [102]\n",
    "\n",
    "# Perform the forward pass and print the logits\n",
    "with torch.no_grad():\n",
    "    print(model(torch.tensor([token_ids])).logits)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[ 4.1692, -3.3464]])\n"
     ]
    }
   ],
   "source": [
    "# Do the entire forward pass manually for sentence 2\n",
    "\n",
    "# Tokenize the sentence and get the tokenids\n",
    "s0_ids = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(sentences[0]))\n",
    "token_ids = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(sentences[1]))\n",
    "s1_tokens = len(token_ids)\n",
    "additional_ids = len(s0_ids) - len(token_ids)\n",
    "\n",
    "# Add the special token CLS and SEP at the start and end of the token repectively\n",
    "# Also create an attention mask here to stop the attention from considering additional padding tokens\n",
    "token_ids = [101] + token_ids + [102] + [0 for _ in range(additional_ids)]\n",
    "attention_mask = [1 for _ in range(s1_tokens + 2)] + [0 for _ in range(additional_ids)]\n",
    "\n",
    "# Perform the forward pass and print the logits\n",
    "with torch.no_grad():\n",
    "    print(model(input_ids = torch.tensor([token_ids]),\n",
    "                attention_mask = torch.tensor([attention_mask])).logits)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}