llama_embed.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Unnamed: 0</th>\n",
       "      <th>Sentence</th>\n",
       "      <th>Label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>The Cardiff Roller Collective (CRoC) are a rol...</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>\"Go! Pack Go!\" is the fight song of the Green ...</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>Al-Machriq (English translation: The East) was...</td>\n",
       "      <td>14</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>Ajman International Airport (Arabic: مطار عجما...</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>Kapla is a construction set for children and a...</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Unnamed: 0                                           Sentence  Label\n",
       "0           0  The Cardiff Roller Collective (CRoC) are a rol...     11\n",
       "1           1  \"Go! Pack Go!\" is the fight song of the Green ...     11\n",
       "2           2  Al-Machriq (English translation: The East) was...     14\n",
       "3           3  Ajman International Airport (Arabic: مطار عجما...      0\n",
       "4           4  Kapla is a construction set for children and a...      4"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv('data.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer, AutoModel\n",
    "import transformers\n",
    "import torch\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = \"meta-llama/Llama-2-7b-chat-hf\"\n",
    "\n",
    "# meta-llama/Meta-Llama-3-8B\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]"
     ]
    }
   ],
   "source": [
    "pipeline = transformers.pipeline(\"feature-extraction\", model=model,torch_dtype=torch.float32)\n",
    "\n",
    "embeds = pipeline(df['Sentence'].tolist(), tokenizer=tokenizer, eos_token=tokenizer.eos_token, padding=True, truncation=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]"
     ]
    }
   ],
   "source": [
    "def get_embedding(text, chunk_size, model_name):\n",
    "    tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
    "    model = AutoModel.from_pretrained(model_name)\n",
    "\n",
    "    tokens = tokenizer(text, return_tensors='pt', padding=True, truncation=False)\n",
    "\n",
    "    # Split tokens to chunks\n",
    "    chunks = [tokens[x:x+chunk_size] for x in range(0, len(tokens), chunk_size)]\n",
    "\n",
    "    # Get embeddings for all chunks\n",
    "    embeddings = []\n",
    "    for chunk in chunks:\n",
    "        with torch.no_grad():\n",
    "            outputs = model(input_ids=chunk.unsqueeze(0))\n",
    "            chunk_embeddings = outputs.last_hidden_state.squeeze(0)\n",
    "            embeddings.append(chunk_embeddings)\n",
    "\n",
    "    # Concatenate all chunks to get the whole sentence embedding\n",
    "    embeddings = torch.cat(embeddings, dim=0)\n",
    "\n",
    "    return embeddings\n",
    "\n",
    "\n",
    "print(get_embedding(df['Sentence'].tolist(), 10, model).shape)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": ".venv",
   "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": 2
}