File size: 9,206 Bytes
7c07f27 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 |
{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"id": "93f5db97-0d94-4464-9891-0ebfe519d534",
"metadata": {},
"outputs": [],
"source": [
"#!pip install -U bitsandbytes\n",
"#!pip install -U transformers\n",
"#!pip install -U accelerate\n",
"#!pip install -U peft\n",
"#!pip install -U trl"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "1fd5f7f5-c053-4ecd-a0d4-b7a12ee32136",
"metadata": {},
"outputs": [],
"source": [
"#!huggingface-cli whoami"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "5780dde2-c61e-464b-91aa-e68301124b6e",
"metadata": {},
"outputs": [],
"source": [
"import numpy as np\n",
"import pandas as pd\n",
"import os\n",
"from tqdm import tqdm\n",
"import bitsandbytes as bnb\n",
"import torch\n",
"import torch.nn as nn\n",
"import transformers\n",
"from datasets import Dataset\n",
"from peft import LoraConfig, PeftConfig\n",
"from trl import SFTTrainer\n",
"from trl import setup_chat_format\n",
"from transformers import (AutoModelForCausalLM, \n",
" AutoTokenizer, \n",
" BitsAndBytesConfig, \n",
" TrainingArguments, \n",
" pipeline, \n",
" logging)\n",
"from sklearn.metrics import (accuracy_score, \n",
" classification_report, \n",
" confusion_matrix)\n",
"from sklearn.model_selection import train_test_split"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "84b29425-b5ad-4852-b9e2-6887eece0de8",
"metadata": {},
"outputs": [],
"source": [
"from datasets import load_dataset\n",
"\n",
"df = pd.read_parquet(\"hf://datasets/tdavidson/hate_speech_offensive/data/train-00000-of-00001.parquet\")\n",
"df.head()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b7395daa-b933-4204-854c-472548343f31",
"metadata": {},
"outputs": [],
"source": [
"df = df.rename(columns={\"class\": \"label\",\"tweet\": \"text\"}).sample(frac=1, random_state=85).reset_index(drop=True).head(3000)\n",
"df.loc[:,'label'] = df.loc[:,'label'].replace(0,'Hate')\n",
"df.loc[:,'label'] = df.loc[:,'label'].replace(1,'Offensive')\n",
"df.loc[:,'label'] = df.loc[:,'label'].replace(2,'Normal')\n",
"# Split the DataFrame\n",
"train_size = 0.8\n",
"eval_size = 0.1\n",
"\n",
"# Calculate sizes\n",
"train_end = int(train_size * len(df))\n",
"eval_end = train_end + int(eval_size * len(df))\n",
"\n",
"# Split the data\n",
"X_train = df[:train_end]\n",
"X_eval = df[train_end:eval_end]\n",
"X_test = df[eval_end:]\n",
"# Define the prompt generation functions\n",
"def generate_prompt(data_point):\n",
" return f\"\"\"\n",
" Classify the text into Hatespeech, Offensive, Normal and return the answer as the corresponding label.\n",
"text: {data_point[\"text\"]}\n",
"label: {data_point[\"label\"]}\"\"\".strip()\n",
"\n",
"def generate_test_prompt(data_point):\n",
" return f\"\"\"\n",
" Classify the text into Hatespeech, Offensive, Normal and return the answer as the corresponding label.\n",
" text: {data_point[\"text\"]}\n",
" label: \"\"\".strip()\n",
"\n",
"# Generate prompts for training and evaluation data\n",
"X_train.loc[:,'text'] = X_train.apply(generate_prompt, axis=1)\n",
"X_eval.loc[:,'text'] = X_eval.apply(generate_prompt, axis=1)\n",
"\n",
"# Generate test prompts and extract true labels\n",
"y_true = X_test.loc[:,'label']\n",
"X_test = pd.DataFrame(X_test.apply(generate_test_prompt, axis=1), columns=[\"text\"])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "bc18edca-e02b-4a32-8cc1-7d83f00bdba5",
"metadata": {},
"outputs": [],
"source": [
"X_train.label.value_counts()"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "52d5ccf5-7669-447f-8a90-43cbb7e8e337",
"metadata": {},
"outputs": [],
"source": [
"train_data = Dataset.from_pandas(X_train[[\"text\"]])\n",
"eval_data = Dataset.from_pandas(X_eval[[\"text\"]])"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f7732c58-e8c6-436b-810d-40abd4f593ab",
"metadata": {},
"outputs": [],
"source": [
"train_data['text'][2000]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "0f15e59f-9e50-48f1-b6f5-d6dc46db623f",
"metadata": {},
"outputs": [],
"source": [
"#CHANGE MODEL HERE#\n",
"base_model_name = \"meta-llama/Llama-3.2-3B-Instruct\"\n",
"\n",
"bnb_config = BitsAndBytesConfig(\n",
" load_in_4bit=True,\n",
" bnb_4bit_use_double_quant=False,\n",
" bnb_4bit_quant_type=\"nf4\",\n",
" bnb_4bit_compute_dtype=\"float16\",\n",
")\n",
"\n",
"model = AutoModelForCausalLM.from_pretrained(\n",
" base_model_name,\n",
" device_map=\"auto\",\n",
" torch_dtype=\"float16\",\n",
" quantization_config=bnb_config, \n",
")\n",
"\n",
"model.config.use_cache = False\n",
"model.config.pretraining_tp = 1\n",
"\n",
"tokenizer = AutoTokenizer.from_pretrained(base_model_name)\n",
"\n",
"tokenizer.pad_token_id = tokenizer.eos_token_id"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "97ccf698-09de-4423-9287-8dedf779fc3d",
"metadata": {},
"outputs": [],
"source": [
"def predict(test, model, tokenizer):\n",
" y_pred = []\n",
" labels = [\"Hate\", \"Offensive\", \"Normal\"]\n",
" \n",
" for i in tqdm(range(len(test))):\n",
" prompt = test.iloc[i][\"text\"]\n",
" pipe = pipeline(task=\"text-generation\", \n",
" model=model, \n",
" tokenizer=tokenizer, \n",
" max_new_tokens=2, \n",
" temperature=0.1)\n",
" \n",
" result = pipe(prompt)\n",
" answer = result[0]['generated_text'].split(\"label:\")[-1].strip()\n",
" \n",
" # Determine the predicted category\n",
" for label in labels:\n",
" if label.lower() in answer.lower():\n",
" y_pred.append(label)\n",
" break\n",
" else:\n",
" y_pred.append(\"none\")\n",
" \n",
" return y_pred\n",
"\n",
"y_pred = predict(X_test, model, tokenizer)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "2bc4f2ea-5cde-4368-8f92-7883995d8977",
"metadata": {},
"outputs": [],
"source": [
"def evaluate(y_true, y_pred):\n",
" labels = [\"Hate\", \"Offensive\", \"Normal\"]\n",
" mapping = {label: idx for idx, label in enumerate(labels)}\n",
" \n",
" def map_func(x):\n",
" return mapping.get(x, -1) # Map to -1 if not found, but should not occur with correct data\n",
" \n",
" y_true_mapped = np.vectorize(map_func)(y_true)\n",
" y_pred_mapped = np.vectorize(map_func)(y_pred)\n",
" \n",
" # Calculate accuracy\n",
" accuracy = accuracy_score(y_true=y_true_mapped, y_pred=y_pred_mapped)\n",
" print(f'Accuracy: {accuracy:.3f}')\n",
" \n",
" # Generate accuracy report\n",
" unique_labels = set(y_true_mapped) # Get unique labels\n",
" \n",
" for label in unique_labels:\n",
" label_indices = [i for i in range(len(y_true_mapped)) if y_true_mapped[i] == label]\n",
" label_y_true = [y_true_mapped[i] for i in label_indices]\n",
" label_y_pred = [y_pred_mapped[i] for i in label_indices]\n",
" label_accuracy = accuracy_score(label_y_true, label_y_pred)\n",
" print(f'Accuracy for label {labels[label]}: {label_accuracy:.3f}')\n",
" \n",
" # Generate classification report\n",
" class_report = classification_report(y_true=y_true_mapped, y_pred=y_pred_mapped, target_names=labels, labels=list(range(len(labels))))\n",
" print('\\nClassification Report:')\n",
" print(class_report)\n",
" \n",
" # Generate confusion matrix\n",
" conf_matrix = confusion_matrix(y_true=y_true_mapped, y_pred=y_pred_mapped, labels=list(range(len(labels))))\n",
" print('\\nConfusion Matrix:')\n",
" print(conf_matrix)\n",
"\n",
"evaluate(y_true, y_pred)"
]
}
],
"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": "3.12.5"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|