File size: 10,642 Bytes
ae3aa18 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 | {
"cells": [
{
"cell_type": "markdown",
"id": "e11600f1",
"metadata": {},
"source": [
"# EmpathRAG — Corpus Annotation (Day 10)\n",
"Annotate all 1,674,369 FAISS chunks with emotion labels using the trained RoBERTa checkpoint. \n",
"**Runtime: A100 GPU. Expected time: ~25 minutes.**\n",
"\n",
"This notebook:\n",
"1. Copies `metadata.db` from your local machine (uploaded here) to Colab\n",
"2. Loads the RoBERTa+LoRA checkpoint from Drive\n",
"3. Runs inference in batches of 512 on GPU\n",
"4. Writes emotion labels back to the SQLite db\n",
"5. Saves the annotated db back to Drive for you to download\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e85f0c1f",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"gpu = torch.cuda.get_device_name(0) if torch.cuda.is_available() else \"NO GPU\"\n",
"print(f\"GPU : {gpu}\")\n",
"print(f\"VRAM : {torch.cuda.get_device_properties(0).total_memory/1e9:.1f} GB\")\n",
"assert torch.cuda.is_available(), \"Switch to A100 runtime.\"\n",
"print(\"✅ GPU ready.\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "90dcada6",
"metadata": {},
"outputs": [],
"source": [
"!pip install -q peft>=0.18.0 accelerate>=1.0.0\n",
"import peft, accelerate\n",
"print(f\"peft: {peft.__version__} | accelerate: {accelerate.__version__}\")\n",
"print(\"✅ Packages ready.\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "51d42310",
"metadata": {},
"outputs": [],
"source": [
"from google.colab import drive\n",
"import os, shutil\n",
"\n",
"drive.mount(\"/content/drive\")\n",
"\n",
"CHECKPOINT_DIR = \"/content/drive/MyDrive/empathrag/emotion_classifier\"\n",
"DB_DRIVE_PATH = \"/content/drive/MyDrive/empathrag/metadata.db\"\n",
"DB_LOCAL_PATH = \"/content/metadata.db\"\n",
"\n",
"# Verify checkpoint exists\n",
"required = [\"adapter_config.json\", \"adapter_model.safetensors\", \"tokenizer.json\"]\n",
"for f in required:\n",
" path = os.path.join(CHECKPOINT_DIR, f)\n",
" assert os.path.exists(path), f\"Missing: {path}\"\n",
"print(f\"✅ Checkpoint found at: {CHECKPOINT_DIR}\")\n"
]
},
{
"cell_type": "markdown",
"id": "f6533428",
"metadata": {},
"source": [
"## ⬆️ Upload metadata.db\n",
"\n",
"Run the cell below. It will prompt you to upload a file. \n",
"Upload `data/indexes/metadata.db` from your local machine.\n",
"\n",
"The file is ~1.26 GB — upload will take 2-5 minutes depending on your connection.\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "8f3928e3",
"metadata": {},
"outputs": [],
"source": [
"from google.colab import files\n",
"import shutil, os\n",
"\n",
"print(\"Click 'Choose Files' and select data/indexes/metadata.db from your local machine.\")\n",
"print(\"File size: ~1.26 GB. Wait for upload to complete before proceeding.\")\n",
"\n",
"uploaded = files.upload()\n",
"\n",
"# Move to working path\n",
"for fname in uploaded:\n",
" dest = \"/content/metadata.db\"\n",
" if fname != \"metadata.db\":\n",
" os.rename(fname, dest)\n",
" else:\n",
" pass # already at /content/metadata.db if uploaded directly\n",
"\n",
"# Verify\n",
"import sqlite3\n",
"conn = sqlite3.connect(DB_LOCAL_PATH)\n",
"total = conn.execute(\"SELECT COUNT(*) FROM chunks\").fetchone()[0]\n",
"annotated = conn.execute(\"SELECT COUNT(*) FROM chunks WHERE emotion_label != -1\").fetchone()[0]\n",
"unannotated= conn.execute(\"SELECT COUNT(*) FROM chunks WHERE emotion_label = -1\").fetchone()[0]\n",
"conn.close()\n",
"\n",
"print(f\"\\nDB loaded:\")\n",
"print(f\" Total rows : {total:,}\")\n",
"print(f\" Annotated : {annotated:,} (already done — will skip)\")\n",
"print(f\" Unannotated : {unannotated:,} (will process these)\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "444f0fea",
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"from transformers import AutoTokenizer, AutoModelForSequenceClassification\n",
"from peft import PeftModel\n",
"\n",
"LABEL_NAMES = [\"distress\", \"anxiety\", \"frustration\", \"neutral\", \"hopeful\"]\n",
"SAFETY_SCORE_MAP = {0: 0.0, 1: 0.0, 2: 0.3, 3: 0.7, 4: 1.0}\n",
"DEVICE = torch.device(\"cuda\")\n",
"\n",
"print(\"Loading RoBERTa + LoRA checkpoint onto GPU...\")\n",
"tokenizer = AutoTokenizer.from_pretrained(CHECKPOINT_DIR)\n",
"base = AutoModelForSequenceClassification.from_pretrained(\"roberta-base\", num_labels=5)\n",
"model = PeftModel.from_pretrained(base, CHECKPOINT_DIR).to(DEVICE).eval()\n",
"\n",
"# Quick sanity check\n",
"enc = tokenizer(\"I feel completely overwhelmed and hopeless\",\n",
" return_tensors=\"pt\", truncation=True, max_length=128).to(DEVICE)\n",
"with torch.no_grad():\n",
" pred = model(**enc).logits.argmax(-1).item()\n",
"print(f\"Sanity check: 'overwhelmed and hopeless' → {LABEL_NAMES[pred]}\")\n",
"assert LABEL_NAMES[pred] == \"distress\", f\"Expected distress, got {LABEL_NAMES[pred]}\"\n",
"print(\"✅ Model loaded and verified on GPU.\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "9e8ef3c2",
"metadata": {},
"outputs": [],
"source": [
"import sqlite3\n",
"from tqdm import tqdm\n",
"\n",
"BATCH_SIZE = 512 # A100 can handle 512 at max_len=128 comfortably\n",
"\n",
"conn = sqlite3.connect(DB_LOCAL_PATH)\n",
"\n",
"# Only fetch rows not yet annotated — resumes safely if interrupted\n",
"rows = conn.execute(\n",
" \"SELECT id, text FROM chunks WHERE emotion_label = -1\"\n",
").fetchall()\n",
"\n",
"print(f\"Rows to annotate: {len(rows):,}\")\n",
"if len(rows) == 0:\n",
" print(\"Nothing to do — all rows already annotated.\")\n",
"else:\n",
" updates = []\n",
"\n",
" for i in tqdm(range(0, len(rows), BATCH_SIZE), desc=\"Annotating\"):\n",
" batch = rows[i : i + BATCH_SIZE]\n",
" ids = [r[0] for r in batch]\n",
" texts = [r[1] for r in batch]\n",
"\n",
" enc = tokenizer(\n",
" texts,\n",
" truncation=True,\n",
" max_length=128,\n",
" padding=True,\n",
" return_tensors=\"pt\"\n",
" ).to(DEVICE)\n",
"\n",
" with torch.no_grad():\n",
" logits = model(**enc).logits\n",
"\n",
" labels = logits.argmax(-1).tolist()\n",
"\n",
" for rid, lbl in zip(ids, labels):\n",
" score = SAFETY_SCORE_MAP[lbl]\n",
" updates.append((lbl, score, rid))\n",
"\n",
" # Commit every 50 batches (~25,600 rows) to avoid losing work if disconnected\n",
" if (i // BATCH_SIZE) % 50 == 0 and updates:\n",
" conn.executemany(\n",
" \"UPDATE chunks SET emotion_label=?, safety_score=? WHERE id=?\",\n",
" updates\n",
" )\n",
" conn.commit()\n",
" updates = []\n",
"\n",
" # Final commit\n",
" if updates:\n",
" conn.executemany(\n",
" \"UPDATE chunks SET emotion_label=?, safety_score=? WHERE id=?\",\n",
" updates\n",
" )\n",
" conn.commit()\n",
"\n",
"print(\"\\n✅ Annotation complete.\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "f434d2b2",
"metadata": {},
"outputs": [],
"source": [
"from collections import Counter\n",
"\n",
"total = conn.execute(\"SELECT COUNT(*) FROM chunks\").fetchone()[0]\n",
"annotated = conn.execute(\"SELECT COUNT(*) FROM chunks WHERE emotion_label != -1\").fetchone()[0]\n",
"unannotated= conn.execute(\"SELECT COUNT(*) FROM chunks WHERE emotion_label = -1\").fetchone()[0]\n",
"\n",
"print(f\"Total rows : {total:,}\")\n",
"print(f\"Annotated : {annotated:,}\")\n",
"print(f\"Unannotated : {unannotated:,} ← must be 0\")\n",
"\n",
"rows = conn.execute(\"SELECT emotion_label FROM chunks\").fetchall()\n",
"dist = Counter(r[0] for r in rows)\n",
"print(\"\\nEmotion distribution:\")\n",
"for i, name in enumerate(LABEL_NAMES):\n",
" pct = 100 * dist[i] / total\n",
" print(f\" {i} {name:<15} {dist[i]:>9,} ({pct:.1f}%)\")\n",
"\n",
"conn.close()\n",
"assert unannotated == 0, f\"{unannotated:,} rows still unannotated!\"\n",
"print(\"\\n✅ All rows annotated.\")\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "945552e5",
"metadata": {},
"outputs": [],
"source": [
"import shutil\n",
"\n",
"print(f\"Copying annotated DB to Drive: {DB_DRIVE_PATH}\")\n",
"os.makedirs(os.path.dirname(DB_DRIVE_PATH), exist_ok=True)\n",
"shutil.copy2(DB_LOCAL_PATH, DB_DRIVE_PATH)\n",
"\n",
"size_mb = os.path.getsize(DB_DRIVE_PATH) / 1e6\n",
"print(f\"Saved: {DB_DRIVE_PATH} ({size_mb:.1f} MB)\")\n",
"print(\"\\n✅ Done. Download metadata.db from Drive and replace data/indexes/metadata.db locally.\")\n"
]
},
{
"cell_type": "markdown",
"id": "ae95d9cd",
"metadata": {},
"source": [
"## ✅ Annotation Complete — Download Steps\n",
"\n",
"1. In Google Drive, navigate to `My Drive/empathrag/metadata.db`\n",
"2. Right-click → Download\n",
"3. Replace your local `data/indexes/metadata.db` with the downloaded file\n",
"4. Verify locally:\n",
"```python\n",
"import sqlite3\n",
"from collections import Counter\n",
"conn = sqlite3.connect('data/indexes/metadata.db')\n",
"unannotated = conn.execute('SELECT COUNT(*) FROM chunks WHERE emotion_label = -1').fetchone()[0]\n",
"print(f'Unannotated: {unannotated}') # must be 0\n",
"conn.close()\n",
"```\n"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"gpuType": "A100",
"name": "colab_annotate_corpus.ipynb",
"provenance": []
},
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"name": "python",
"version": "3.12.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|