added bloomz query notebook

jupyter_notebooks/.ipynb_checkpoints/Section_2-3-4_Bloomz_query-checkpoint.ipynb

@@ -0,0 +1,370 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "3b87c378-241e-41ab-be6e-84222594f22f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "import json\n",
+    "import time\n",
+    "import re\n",
+    "from pathlib import Path\n",
+    "from tqdm import tqdm\n",
+    "import torch\n",
+    "from transformers import AutoModelForCausalLM, AutoTokenizer\n",
+    "\n",
+    "# Import is used for pd.notna() and pd.isna() checks\n",
+    "\n",
+    "current_dir = Path.cwd()\n",
+    "input_file = current_dir.parent / \"data/CSV/model_adapter/real_person_adapter_step_02_NER.csv\"\n",
+    "\n",
+    "# === CONFIGURATION ===\n",
+    "TEST_MODE = True\n",
+    "TEST_SIZE = 10\n",
+    "MAX_ROWS = 20000\n",
+    "SAVE_INTERVAL = 10\n",
+    "\n",
+    "# Model settings - BLOOMZ (BigScience - European consortium)\n",
+    "MODEL_NAME = \"bigscience/bloomz-7b1\"  # Largest instruction-tuned BLOOM model\n",
+    "CACHE_DIR = current_dir.parent / \"data/models\"\n",
+    "CACHE_DIR.mkdir(parents=True, exist_ok=True)\n",
+    "\n",
+    "PROFESSION_CATEGORIES = [\n",
+    "    \"actor\", \"adult performer\", \"singer/musician\", \"model\",\n",
+    "    \"online personality\", \"public figure\", \"voice actor/ASMR\",\n",
+    "    \"sports professional\", \"tv personality\"\n",
+    "]\n",
+    "\n",
+    "# === LOAD MODEL ===\n",
+    "print(f\"Loading model: {MODEL_NAME}\")\n",
+    "print(f\"Cache directory: {CACHE_DIR}\")\n",
+    "print(f\"This may take a while on first run (~14GB download)...\\n\")\n",
+    "\n",
+    "# Check GPU availability\n",
+    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "print(f\"Device: {device}\")\n",
+    "\n",
+    "if device == \"cpu\":\n",
+    "    print(\"⚠️  WARNING: No GPU detected! Inference will be VERY slow.\")\n",
+    "    print(\"   Consider using a GPU or reducing model size.\")\n",
+    "\n",
+    "# Load tokenizer\n",
+    "print(\"Loading tokenizer...\")\n",
+    "try:\n",
+    "    tokenizer = AutoTokenizer.from_pretrained(\n",
+    "        MODEL_NAME,\n",
+    "        cache_dir=str(CACHE_DIR)\n",
+    "    )\n",
+    "    print(\"✅ Tokenizer loaded\")\n",
+    "except Exception as e:\n",
+    "    print(f\"❌ Error loading tokenizer: {e}\")\n",
+    "    raise\n",
+    "\n",
+    "# Ensure pad token is set\n",
+    "if tokenizer.pad_token is None:\n",
+    "    tokenizer.pad_token = tokenizer.eos_token\n",
+    "    print(f\"Set pad_token to eos_token: {tokenizer.eos_token}\")\n",
+    "\n",
+    "# Load model with optimizations\n",
+    "print(\"Loading model (this may take several minutes)...\")\n",
+    "try:\n",
+    "    model = AutoModelForCausalLM.from_pretrained(\n",
+    "        MODEL_NAME,\n",
+    "        cache_dir=str(CACHE_DIR),\n",
+    "        torch_dtype=torch.bfloat16,  # Use BF16 for efficiency\n",
+    "        device_map=\"auto\",  # Automatically distribute across GPUs\n",
+    "        low_cpu_mem_usage=True  # Optimize memory usage\n",
+    "    )\n",
+    "    model.eval()  # Set to evaluation mode\n",
+    "    print(\"✅ Model loaded\")\n",
+    "except Exception as e:\n",
+    "    print(f\"❌ Error loading model: {e}\")\n",
+    "    raise\n",
+    "\n",
+    "# Check VRAM usage\n",
+    "if torch.cuda.is_available():\n",
+    "    vram_gb = torch.cuda.max_memory_allocated() / 1024**3\n",
+    "    print(f\"VRAM used: {vram_gb:.2f} GB\\n\")\n",
+    "\n",
+    "# === LOAD DATA ===\n",
+    "df = pd.read_csv(input_file)\n",
+    "print(f\"Loaded {len(df)} rows\")\n",
+    "\n",
+    "if TEST_MODE:\n",
+    "    print(f\"Running in TEST MODE with {TEST_SIZE} samples\")\n",
+    "    df = df.head(TEST_SIZE).copy()\n",
+    "elif MAX_ROWS:\n",
+    "    df = df.head(MAX_ROWS).copy()\n",
+    "\n",
+    "# === CREATE PROMPT (Exact DeepSeek style) ===\n",
+    "def create_prompt(row):\n",
+    "    \"\"\"Create prompt.\"\"\"\n",
+    "    name = row.get('real_name', row.get('name', ''))\n",
+    "    if pd.isna(name):\n",
+    "        name = row.get('name', '')\n",
+    "    \n",
+    "    # Gather hints exactly like DeepSeek version\n",
+    "    hints = []\n",
+    "    if pd.notna(row.get('likely_profession')):\n",
+    "        hints.append(str(row['likely_profession']))\n",
+    "    if pd.notna(row.get('likely_nationality')):\n",
+    "        hints.append(str(row['likely_nationality']))\n",
+    "    if pd.notna(row.get('likely_country')):\n",
+    "        hints.append(str(row['likely_country']))\n",
+    "    \n",
+    "    # Add tags if we don't have enough hints\n",
+    "    if len(hints) < 3:\n",
+    "        for i in range(1, 8):\n",
+    "            tag_col = f'tag_{i}'\n",
+    "            if tag_col in row and pd.notna(row[tag_col]):\n",
+    "                tag_val = str(row[tag_col])\n",
+    "                if tag_val not in hints:\n",
+    "                    hints.append(tag_val)\n",
+    "                if len(hints) >= 5:\n",
+    "                    break\n",
+    "    \n",
+    "    hint_text = \", \".join(hints[:5]) if hints else \"none\"\n",
+    "    \n",
+    "    return f\"\"\"Given '{name}' ({hint_text}), provide:\n",
+    "1. Full legal name (Western order if non-latin script)\n",
+    "2. Any stage names/aliases (comma separated)\n",
+    "3. Gender (Male/Female/Other/Unknown)\n",
+    "4. Top 3 most likely professions from ONLY these categories:\n",
+    "   - actor\n",
+    "   - adult performer\n",
+    "   - singer/musician\n",
+    "   - model\n",
+    "   - online personality (includes streamers, cosplayers, influencers)\n",
+    "   - public figure (includes politicians, activists, journalists, authors)\n",
+    "   - voice actor/ASMR\n",
+    "   - sports professional\n",
+    "   - tv personality (includes hosts, presenters, reality TV)\n",
+    "\n",
+    "5. Primary country associated\n",
+    "\n",
+    "IMPORTANT:\n",
+    "- Choose professions ONLY from the 9 categories above\n",
+    "- Provide up to 3 professions, comma-separated, ordered by relevance\n",
+    "- Be SPECIFIC: choose the most accurate category for each role\n",
+    "- \"online personality\" includes: streamers, cosplayers, YouTubers, influencers, content creators\n",
+    "- Use 'Unknown' when uncertain or for fictional characters/places\n",
+    "- For multi-role people, list all relevant categories (e.g., \"actor, singer/musician, online personality\")\n",
+    "\n",
+    "Respond with exactly 5 numbered lines.\"\"\"\n",
+    "\n",
+    "df['prompt'] = df.apply(create_prompt, axis=1)\n",
+    "\n",
+    "# === QUERY BLOOMZ LOCAL ===\n",
+    "def query_bloomz_local(prompt: str) -> str:\n",
+    "    \"\"\"Query BLOOMZ-7B1 locally via transformers, return raw response string.\"\"\"\n",
+    "    try:\n",
+    "        # BLOOMZ works better with instruction-response format\n",
+    "        full_prompt = f\"\"\"Instruction: Extract key data on a person based on the name and hints.\n",
+    "You must respond with exactly 5 numbered lines in this format:\n",
+    "1. Full legal name\n",
+    "2. Stage names/aliases  \n",
+    "3. Gender\n",
+    "4. Professions (comma-separated, choose ONLY from: actor, adult performer, singer/musician, model, online personality, public figure, voice actor/ASMR, sports professional, tv personality)\n",
+    "5. Country\n",
+    "\n",
+    "{prompt}\n",
+    "\n",
+    "Response:\"\"\"\n",
+    "        \n",
+    "        inputs = tokenizer(\n",
+    "            full_prompt, \n",
+    "            return_tensors=\"pt\", \n",
+    "            truncation=True,\n",
+    "            max_length=2048\n",
+    "        ).to(device)\n",
+    "        \n",
+    "        # Generate with adjusted parameters for BLOOMZ\n",
+    "        with torch.no_grad():\n",
+    "            outputs = model.generate(\n",
+    "                **inputs,\n",
+    "                max_new_tokens=256,\n",
+    "                temperature=0.3,  # Increased for more variability\n",
+    "                do_sample=True,\n",
+    "                top_p=0.9,\n",
+    "                top_k=40,\n",
+    "                repetition_penalty=1.1,\n",
+    "                pad_token_id=tokenizer.eos_token_id,  # Use EOS as pad token\n",
+    "                eos_token_id=tokenizer.eos_token_id,\n",
+    "                early_stopping=True\n",
+    "            )\n",
+    "        \n",
+    "        # Decode the entire output to see what's happening\n",
+    "        full_output = tokenizer.decode(outputs[0], skip_special_tokens=True)\n",
+    "        \n",
+    "        # Extract only the generated part (after the prompt)\n",
+    "        generated_text = full_output[len(tokenizer.decode(inputs['input_ids'][0], skip_special_tokens=True)):]\n",
+    "        \n",
+    "        # Debug output\n",
+    "        if not hasattr(query_bloomz_local, 'debug_count'):\n",
+    "            query_bloomz_local.debug_count = 0\n",
+    "        \n",
+    "        if query_bloomz_local.debug_count < 3:\n",
+    "            print(f\"\\n📝 BLOOMZ Debug #{query_bloomz_local.debug_count + 1}:\")\n",
+    "            print(f\"Prompt: {full_prompt[:200]}...\")\n",
+    "            print(f\"Full output: {full_output[:500]}...\")\n",
+    "            print(f\"Generated text: {generated_text}\")\n",
+    "            print(f\"{'='*60}\\n\")\n",
+    "            query_bloomz_local.debug_count += 1\n",
+    "        \n",
+    "        return generated_text.strip()\n",
+    "        \n",
+    "    except Exception as e:\n",
+    "        print(f\"Error querying BLOOMZ: {e}\")\n",
+    "        return None\n",
+    "\n",
+    "# === PARSE RESPONSE (Exact DeepSeek format) ===\n",
+    "def parse_response(response):\n",
+    "    \"\"\"Parse numbered response into structured fields.\"\"\"\n",
+    "    if not response:\n",
+    "        return {\n",
+    "            'full_name': 'Unknown',\n",
+    "            'aliases': 'Unknown',\n",
+    "            'gender': 'Unknown',\n",
+    "            'profession_llm': 'Unknown',\n",
+    "            'country': 'Unknown'\n",
+    "        }\n",
+    "    \n",
+    "    # Split into lines and clean\n",
+    "    lines = [line.strip() for line in response.split('\\n') if line.strip()]\n",
+    "    \n",
+    "    # Initialize with Unknown values\n",
+    "    fields = {\n",
+    "        'full_name': 'Unknown',\n",
+    "        'aliases': 'Unknown',\n",
+    "        'gender': 'Unknown',\n",
+    "        'profession_llm': 'Unknown',\n",
+    "        'country': 'Unknown'\n",
+    "    }\n",
+    "    \n",
+    "    # Extract information from each numbered line\n",
+    "    for line in lines:\n",
+    "        if line.startswith('1.') or line.startswith('1)'):\n",
+    "            fields['full_name'] = line[2:].strip()\n",
+    "        elif line.startswith('2.') or line.startswith('2)'):\n",
+    "            fields['aliases'] = line[2:].strip()\n",
+    "        elif line.startswith('3.') or line.startswith('3)'):\n",
+    "            fields['gender'] = line[2:].strip()\n",
+    "        elif line.startswith('4.') or line.startswith('4)'):\n",
+    "            fields['profession_llm'] = line[2:].strip()\n",
+    "        elif line.startswith('5.') or line.startswith('5)'):\n",
+    "            fields['country'] = line[2:].strip()\n",
+    "    \n",
+    "    return fields\n",
+    "\n",
+    "# === PROCESS ===\n",
+    "output_file = current_dir.parent / f\"data/CSV/bloomz_annotated_POI{'_test' if TEST_MODE else ''}.csv\"\n",
+    "index_file = current_dir.parent / \"misc/bloomz_query_index.txt\"\n",
+    "\n",
+    "current_index = 0\n",
+    "if index_file.exists():\n",
+    "    with open(index_file) as f:\n",
+    "        current_index = int(f.read().strip())\n",
+    "    print(f\"Resuming from index {current_index}\")\n",
+    "\n",
+    "# Initialize columns (same as DeepSeek)\n",
+    "for col in ['full_name', 'gender', 'profession_llm', 'country', 'aliases']:\n",
+    "    if col not in df.columns:\n",
+    "        df[col] = 'Unknown'\n",
+    "\n",
+    "# Create prompts for all rows (same as DeepSeek)\n",
+    "print(\"Creating prompts...\")\n",
+    "df['prompt'] = df.apply(create_prompt, axis=1)\n",
+    "\n",
+    "print(f\"\\nAnnotating with BLOOMZ-7B1 LOCAL - rows {current_index} to {len(df)}...\")\n",
+    "print(f\"Model: {MODEL_NAME}\")\n",
+    "print(f\"This may take a while...\\n\")\n",
+    "\n",
+    "try:\n",
+    "    start_time = time.time()\n",
+    "    \n",
+    "    for i in tqdm(range(current_index, len(df)), desc=\"Annotating\"):\n",
+    "        row = df.iloc[i]\n",
+    "        \n",
+    "        # Query BLOOMZ (equivalent to DeepSeek query)\n",
+    "        response = query_bloomz_local(row['prompt'])\n",
+    "        parsed_data = parse_response(response)\n",
+    "        \n",
+    "        # Update dataframe\n",
+    "        for key, value in parsed_data.items():\n",
+    "            df.at[i, key] = value\n",
+    "        \n",
+    "        current_index = i + 1\n",
+    "        \n",
+    "        # Save progress at intervals\n",
+    "        if (i + 1) % SAVE_INTERVAL == 0 or (i + 1) == len(df):\n",
+    "            df.to_csv(output_file, index=False)\n",
+    "            with open(index_file, 'w') as f:\n",
+    "                f.write(str(current_index))\n",
+    "            print(f\"✅ Progress saved after {i+1} rows\")\n",
+    "        \n",
+    "        # Optional: Add small delay to prevent overheating (not needed for rate limiting like DeepSeek)\n",
+    "        # time.sleep(0.1)\n",
+    "    \n",
+    "    elapsed_total = time.time() - start_time\n",
+    "    print(f\"\\n✅ Done! Final results saved to {output_file}\")\n",
+    "    \n",
+    "    # Summary statistics (same as DeepSeek)\n",
+    "    print(\"\\n=== Summary Statistics ===\")\n",
+    "    print(f\"Total processed: {len(df)}\")\n",
+    "    print(f\"\\nGender distribution:\")\n",
+    "    print(df['gender'].value_counts())\n",
+    "    print(f\"\\nTop 10 profession combinations:\")\n",
+    "    print(df['profession_llm'].value_counts().head(10))\n",
+    "    print(f\"\\nTop 10 countries:\")\n",
+    "    print(df['country'].value_counts().head(10))\n",
+    "    \n",
+    "    # Sample results\n",
+    "    print(\"\\n=== Sample Results ===\")\n",
+    "    display_cols = ['real_name', 'full_name', 'gender', 'profession_llm', 'country']\n",
+    "    available_cols = [col for col in display_cols if col in df.columns]\n",
+    "    print(df[available_cols].head(10).to_string(index=False))\n",
+    "    \n",
+    "    # Additional info for local model\n",
+    "    print(f\"\\nTotal time: {elapsed_total/60:.1f} minutes\")\n",
+    "    print(f\"Average speed: {len(df)/(elapsed_total/3600):.1f} samples/hour\")\n",
+    "    if torch.cuda.is_available():\n",
+    "        print(f\"Final VRAM usage: {torch.cuda.max_memory_allocated() / 1024**3:.2f} GB\")\n",
+    "\n",
+    "except Exception as e:\n",
+    "    print(f\"⚠️ Error encountered: {e}\")\n",
+    "    print(f\"⚠️ Last processed index: {current_index}\")\n",
+    "    \n",
+    "    # Save progress before exiting\n",
+    "    df.to_csv(output_file, index=False)\n",
+    "    with open(index_file, 'w') as f:\n",
+    "        f.write(str(current_index))\n",
+    "    \n",
+    "    print(f\"⚠️ Progress saved up to row {current_index}\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "pm-paper",
+   "language": "python",
+   "name": "pm-paper"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

jupyter_notebooks/Section_2-3-4_Bloomz_query.ipynb

@@ -0,0 +1,370 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "3b87c378-241e-41ab-be6e-84222594f22f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "import json\n",
+    "import time\n",
+    "import re\n",
+    "from pathlib import Path\n",
+    "from tqdm import tqdm\n",
+    "import torch\n",
+    "from transformers import AutoModelForCausalLM, AutoTokenizer\n",
+    "\n",
+    "# Import is used for pd.notna() and pd.isna() checks\n",
+    "\n",
+    "current_dir = Path.cwd()\n",
+    "input_file = current_dir.parent / \"data/CSV/model_adapter/real_person_adapter_step_02_NER.csv\"\n",
+    "\n",
+    "# === CONFIGURATION ===\n",
+    "TEST_MODE = True\n",
+    "TEST_SIZE = 10\n",
+    "MAX_ROWS = 20000\n",
+    "SAVE_INTERVAL = 10\n",
+    "\n",
+    "# Model settings - BLOOMZ (BigScience - European consortium)\n",
+    "MODEL_NAME = \"bigscience/bloomz-7b1\"  # Largest instruction-tuned BLOOM model\n",
+    "CACHE_DIR = current_dir.parent / \"data/models\"\n",
+    "CACHE_DIR.mkdir(parents=True, exist_ok=True)\n",
+    "\n",
+    "PROFESSION_CATEGORIES = [\n",
+    "    \"actor\", \"adult performer\", \"singer/musician\", \"model\",\n",
+    "    \"online personality\", \"public figure\", \"voice actor/ASMR\",\n",
+    "    \"sports professional\", \"tv personality\"\n",
+    "]\n",
+    "\n",
+    "# === LOAD MODEL ===\n",
+    "print(f\"Loading model: {MODEL_NAME}\")\n",
+    "print(f\"Cache directory: {CACHE_DIR}\")\n",
+    "print(f\"This may take a while on first run (~14GB download)...\\n\")\n",
+    "\n",
+    "# Check GPU availability\n",
+    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "print(f\"Device: {device}\")\n",
+    "\n",
+    "if device == \"cpu\":\n",
+    "    print(\"⚠️  WARNING: No GPU detected! Inference will be VERY slow.\")\n",
+    "    print(\"   Consider using a GPU or reducing model size.\")\n",
+    "\n",
+    "# Load tokenizer\n",
+    "print(\"Loading tokenizer...\")\n",
+    "try:\n",
+    "    tokenizer = AutoTokenizer.from_pretrained(\n",
+    "        MODEL_NAME,\n",
+    "        cache_dir=str(CACHE_DIR)\n",
+    "    )\n",
+    "    print(\"✅ Tokenizer loaded\")\n",
+    "except Exception as e:\n",
+    "    print(f\"❌ Error loading tokenizer: {e}\")\n",
+    "    raise\n",
+    "\n",
+    "# Ensure pad token is set\n",
+    "if tokenizer.pad_token is None:\n",
+    "    tokenizer.pad_token = tokenizer.eos_token\n",
+    "    print(f\"Set pad_token to eos_token: {tokenizer.eos_token}\")\n",
+    "\n",
+    "# Load model with optimizations\n",
+    "print(\"Loading model (this may take several minutes)...\")\n",
+    "try:\n",
+    "    model = AutoModelForCausalLM.from_pretrained(\n",
+    "        MODEL_NAME,\n",
+    "        cache_dir=str(CACHE_DIR),\n",
+    "        torch_dtype=torch.bfloat16,  # Use BF16 for efficiency\n",
+    "        device_map=\"auto\",  # Automatically distribute across GPUs\n",
+    "        low_cpu_mem_usage=True  # Optimize memory usage\n",
+    "    )\n",
+    "    model.eval()  # Set to evaluation mode\n",
+    "    print(\"✅ Model loaded\")\n",
+    "except Exception as e:\n",
+    "    print(f\"❌ Error loading model: {e}\")\n",
+    "    raise\n",
+    "\n",
+    "# Check VRAM usage\n",
+    "if torch.cuda.is_available():\n",
+    "    vram_gb = torch.cuda.max_memory_allocated() / 1024**3\n",
+    "    print(f\"VRAM used: {vram_gb:.2f} GB\\n\")\n",
+    "\n",
+    "# === LOAD DATA ===\n",
+    "df = pd.read_csv(input_file)\n",
+    "print(f\"Loaded {len(df)} rows\")\n",
+    "\n",
+    "if TEST_MODE:\n",
+    "    print(f\"Running in TEST MODE with {TEST_SIZE} samples\")\n",
+    "    df = df.head(TEST_SIZE).copy()\n",
+    "elif MAX_ROWS:\n",
+    "    df = df.head(MAX_ROWS).copy()\n",
+    "\n",
+    "# === CREATE PROMPT (Exact DeepSeek style) ===\n",
+    "def create_prompt(row):\n",
+    "    \"\"\"Create prompt.\"\"\"\n",
+    "    name = row.get('real_name', row.get('name', ''))\n",
+    "    if pd.isna(name):\n",
+    "        name = row.get('name', '')\n",
+    "    \n",
+    "    # Gather hints exactly like DeepSeek version\n",
+    "    hints = []\n",
+    "    if pd.notna(row.get('likely_profession')):\n",
+    "        hints.append(str(row['likely_profession']))\n",
+    "    if pd.notna(row.get('likely_nationality')):\n",
+    "        hints.append(str(row['likely_nationality']))\n",
+    "    if pd.notna(row.get('likely_country')):\n",
+    "        hints.append(str(row['likely_country']))\n",
+    "    \n",
+    "    # Add tags if we don't have enough hints\n",
+    "    if len(hints) < 3:\n",
+    "        for i in range(1, 8):\n",
+    "            tag_col = f'tag_{i}'\n",
+    "            if tag_col in row and pd.notna(row[tag_col]):\n",
+    "                tag_val = str(row[tag_col])\n",
+    "                if tag_val not in hints:\n",
+    "                    hints.append(tag_val)\n",
+    "                if len(hints) >= 5:\n",
+    "                    break\n",
+    "    \n",
+    "    hint_text = \", \".join(hints[:5]) if hints else \"none\"\n",
+    "    \n",
+    "    return f\"\"\"Given '{name}' ({hint_text}), provide:\n",
+    "1. Full legal name (Western order if non-latin script)\n",
+    "2. Any stage names/aliases (comma separated)\n",
+    "3. Gender (Male/Female/Other/Unknown)\n",
+    "4. Top 3 most likely professions from ONLY these categories:\n",
+    "   - actor\n",
+    "   - adult performer\n",
+    "   - singer/musician\n",
+    "   - model\n",
+    "   - online personality (includes streamers, cosplayers, influencers)\n",
+    "   - public figure (includes politicians, activists, journalists, authors)\n",
+    "   - voice actor/ASMR\n",
+    "   - sports professional\n",
+    "   - tv personality (includes hosts, presenters, reality TV)\n",
+    "\n",
+    "5. Primary country associated\n",
+    "\n",
+    "IMPORTANT:\n",
+    "- Choose professions ONLY from the 9 categories above\n",
+    "- Provide up to 3 professions, comma-separated, ordered by relevance\n",
+    "- Be SPECIFIC: choose the most accurate category for each role\n",
+    "- \"online personality\" includes: streamers, cosplayers, YouTubers, influencers, content creators\n",
+    "- Use 'Unknown' when uncertain or for fictional characters/places\n",
+    "- For multi-role people, list all relevant categories (e.g., \"actor, singer/musician, online personality\")\n",
+    "\n",
+    "Respond with exactly 5 numbered lines.\"\"\"\n",
+    "\n",
+    "df['prompt'] = df.apply(create_prompt, axis=1)\n",
+    "\n",
+    "# === QUERY BLOOMZ LOCAL ===\n",
+    "def query_bloomz_local(prompt: str) -> str:\n",
+    "    \"\"\"Query BLOOMZ-7B1 locally via transformers, return raw response string.\"\"\"\n",
+    "    try:\n",
+    "        # BLOOMZ works better with instruction-response format\n",
+    "        full_prompt = f\"\"\"Instruction: Extract key data on a person based on the name and hints.\n",
+    "You must respond with exactly 5 numbered lines in this format:\n",
+    "1. Full legal name\n",
+    "2. Stage names/aliases  \n",
+    "3. Gender\n",
+    "4. Professions (comma-separated, choose ONLY from: actor, adult performer, singer/musician, model, online personality, public figure, voice actor/ASMR, sports professional, tv personality)\n",
+    "5. Country\n",
+    "\n",
+    "{prompt}\n",
+    "\n",
+    "Response:\"\"\"\n",
+    "        \n",
+    "        inputs = tokenizer(\n",
+    "            full_prompt, \n",
+    "            return_tensors=\"pt\", \n",
+    "            truncation=True,\n",
+    "            max_length=2048\n",
+    "        ).to(device)\n",
+    "        \n",
+    "        # Generate with adjusted parameters for BLOOMZ\n",
+    "        with torch.no_grad():\n",
+    "            outputs = model.generate(\n",
+    "                **inputs,\n",
+    "                max_new_tokens=256,\n",
+    "                temperature=0.3,  # Increased for more variability\n",
+    "                do_sample=True,\n",
+    "                top_p=0.9,\n",
+    "                top_k=40,\n",
+    "                repetition_penalty=1.1,\n",
+    "                pad_token_id=tokenizer.eos_token_id,  # Use EOS as pad token\n",
+    "                eos_token_id=tokenizer.eos_token_id,\n",
+    "                early_stopping=True\n",
+    "            )\n",
+    "        \n",
+    "        # Decode the entire output to see what's happening\n",
+    "        full_output = tokenizer.decode(outputs[0], skip_special_tokens=True)\n",
+    "        \n",
+    "        # Extract only the generated part (after the prompt)\n",
+    "        generated_text = full_output[len(tokenizer.decode(inputs['input_ids'][0], skip_special_tokens=True)):]\n",
+    "        \n",
+    "        # Debug output\n",
+    "        if not hasattr(query_bloomz_local, 'debug_count'):\n",
+    "            query_bloomz_local.debug_count = 0\n",
+    "        \n",
+    "        if query_bloomz_local.debug_count < 3:\n",
+    "            print(f\"\\n📝 BLOOMZ Debug #{query_bloomz_local.debug_count + 1}:\")\n",
+    "            print(f\"Prompt: {full_prompt[:200]}...\")\n",
+    "            print(f\"Full output: {full_output[:500]}...\")\n",
+    "            print(f\"Generated text: {generated_text}\")\n",
+    "            print(f\"{'='*60}\\n\")\n",
+    "            query_bloomz_local.debug_count += 1\n",
+    "        \n",
+    "        return generated_text.strip()\n",
+    "        \n",
+    "    except Exception as e:\n",
+    "        print(f\"Error querying BLOOMZ: {e}\")\n",
+    "        return None\n",
+    "\n",
+    "# === PARSE RESPONSE (Exact DeepSeek format) ===\n",
+    "def parse_response(response):\n",
+    "    \"\"\"Parse numbered response into structured fields.\"\"\"\n",
+    "    if not response:\n",
+    "        return {\n",
+    "            'full_name': 'Unknown',\n",
+    "            'aliases': 'Unknown',\n",
+    "            'gender': 'Unknown',\n",
+    "            'profession_llm': 'Unknown',\n",
+    "            'country': 'Unknown'\n",
+    "        }\n",
+    "    \n",
+    "    # Split into lines and clean\n",
+    "    lines = [line.strip() for line in response.split('\\n') if line.strip()]\n",
+    "    \n",
+    "    # Initialize with Unknown values\n",
+    "    fields = {\n",
+    "        'full_name': 'Unknown',\n",
+    "        'aliases': 'Unknown',\n",
+    "        'gender': 'Unknown',\n",
+    "        'profession_llm': 'Unknown',\n",
+    "        'country': 'Unknown'\n",
+    "    }\n",
+    "    \n",
+    "    # Extract information from each numbered line\n",
+    "    for line in lines:\n",
+    "        if line.startswith('1.') or line.startswith('1)'):\n",
+    "            fields['full_name'] = line[2:].strip()\n",
+    "        elif line.startswith('2.') or line.startswith('2)'):\n",
+    "            fields['aliases'] = line[2:].strip()\n",
+    "        elif line.startswith('3.') or line.startswith('3)'):\n",
+    "            fields['gender'] = line[2:].strip()\n",
+    "        elif line.startswith('4.') or line.startswith('4)'):\n",
+    "            fields['profession_llm'] = line[2:].strip()\n",
+    "        elif line.startswith('5.') or line.startswith('5)'):\n",
+    "            fields['country'] = line[2:].strip()\n",
+    "    \n",
+    "    return fields\n",
+    "\n",
+    "# === PROCESS ===\n",
+    "output_file = current_dir.parent / f\"data/CSV/bloomz_annotated_POI{'_test' if TEST_MODE else ''}.csv\"\n",
+    "index_file = current_dir.parent / \"misc/bloomz_query_index.txt\"\n",
+    "\n",
+    "current_index = 0\n",
+    "if index_file.exists():\n",
+    "    with open(index_file) as f:\n",
+    "        current_index = int(f.read().strip())\n",
+    "    print(f\"Resuming from index {current_index}\")\n",
+    "\n",
+    "# Initialize columns (same as DeepSeek)\n",
+    "for col in ['full_name', 'gender', 'profession_llm', 'country', 'aliases']:\n",
+    "    if col not in df.columns:\n",
+    "        df[col] = 'Unknown'\n",
+    "\n",
+    "# Create prompts for all rows (same as DeepSeek)\n",
+    "print(\"Creating prompts...\")\n",
+    "df['prompt'] = df.apply(create_prompt, axis=1)\n",
+    "\n",
+    "print(f\"\\nAnnotating with BLOOMZ-7B1 LOCAL - rows {current_index} to {len(df)}...\")\n",
+    "print(f\"Model: {MODEL_NAME}\")\n",
+    "print(f\"This may take a while...\\n\")\n",
+    "\n",
+    "try:\n",
+    "    start_time = time.time()\n",
+    "    \n",
+    "    for i in tqdm(range(current_index, len(df)), desc=\"Annotating\"):\n",
+    "        row = df.iloc[i]\n",
+    "        \n",
+    "        # Query BLOOMZ (equivalent to DeepSeek query)\n",
+    "        response = query_bloomz_local(row['prompt'])\n",
+    "        parsed_data = parse_response(response)\n",
+    "        \n",
+    "        # Update dataframe\n",
+    "        for key, value in parsed_data.items():\n",
+    "            df.at[i, key] = value\n",
+    "        \n",
+    "        current_index = i + 1\n",
+    "        \n",
+    "        # Save progress at intervals\n",
+    "        if (i + 1) % SAVE_INTERVAL == 0 or (i + 1) == len(df):\n",
+    "            df.to_csv(output_file, index=False)\n",
+    "            with open(index_file, 'w') as f:\n",
+    "                f.write(str(current_index))\n",
+    "            print(f\"✅ Progress saved after {i+1} rows\")\n",
+    "        \n",
+    "        # Optional: Add small delay to prevent overheating (not needed for rate limiting like DeepSeek)\n",
+    "        # time.sleep(0.1)\n",
+    "    \n",
+    "    elapsed_total = time.time() - start_time\n",
+    "    print(f\"\\n✅ Done! Final results saved to {output_file}\")\n",
+    "    \n",
+    "    # Summary statistics (same as DeepSeek)\n",
+    "    print(\"\\n=== Summary Statistics ===\")\n",
+    "    print(f\"Total processed: {len(df)}\")\n",
+    "    print(f\"\\nGender distribution:\")\n",
+    "    print(df['gender'].value_counts())\n",
+    "    print(f\"\\nTop 10 profession combinations:\")\n",
+    "    print(df['profession_llm'].value_counts().head(10))\n",
+    "    print(f\"\\nTop 10 countries:\")\n",
+    "    print(df['country'].value_counts().head(10))\n",
+    "    \n",
+    "    # Sample results\n",
+    "    print(\"\\n=== Sample Results ===\")\n",
+    "    display_cols = ['real_name', 'full_name', 'gender', 'profession_llm', 'country']\n",
+    "    available_cols = [col for col in display_cols if col in df.columns]\n",
+    "    print(df[available_cols].head(10).to_string(index=False))\n",
+    "    \n",
+    "    # Additional info for local model\n",
+    "    print(f\"\\nTotal time: {elapsed_total/60:.1f} minutes\")\n",
+    "    print(f\"Average speed: {len(df)/(elapsed_total/3600):.1f} samples/hour\")\n",
+    "    if torch.cuda.is_available():\n",
+    "        print(f\"Final VRAM usage: {torch.cuda.max_memory_allocated() / 1024**3:.2f} GB\")\n",
+    "\n",
+    "except Exception as e:\n",
+    "    print(f\"⚠️ Error encountered: {e}\")\n",
+    "    print(f\"⚠️ Last processed index: {current_index}\")\n",
+    "    \n",
+    "    # Save progress before exiting\n",
+    "    df.to_csv(output_file, index=False)\n",
+    "    with open(index_file, 'w') as f:\n",
+    "        f.write(str(current_index))\n",
+    "    \n",
+    "    print(f\"⚠️ Progress saved up to row {current_index}\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "pm-paper",
+   "language": "python",
+   "name": "pm-paper"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.13"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

jupyter_notebooks/Section_2-3-4_Figure_8_deepfake_adapters.ipynb

@@ -1829,15 +1829,26 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "id": "72c468dc-1cbe-41d4-b71b-17c40ea4f9a1",
    "metadata": {
     "execution": {
-     "iopub.execute_input": "2025-11-21T19:50:12.690047Z",
-     "iopub.status.busy": "2025-11-21T19:50:12.689854Z"
+     "iopub.execute_input": "2025-11-28T09:54:50.978895Z",
+     "iopub.status.busy": "2025-11-28T09:54:50.978724Z",
+     "iopub.status.idle": "2025-11-28T10:02:38.826002Z",
+     "shell.execute_reply": "2025-11-28T10:02:38.825201Z",
+     "shell.execute_reply.started": "2025-11-28T09:54:50.978881Z"
     }
    },
    "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/shares/weddigen.ki.uzh/laura_wagner/phase_01/pm-paper/.venv/lib/python3.11/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"
+     ]
+    },
     {
      "name": "stdout",
      "output_type": "stream",
@@ -1849,9 +1860,22 @@
       "Device: cuda\n",
       "Loading tokenizer...\n",
       "✅ Tokenizer loaded\n",
-      "Loading model (this may take several minutes)...\n",
+      "Loading model (this may take several minutes)...\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "`torch_dtype` is deprecated! Use `dtype` instead!\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
       "✅ Model loaded\n",
-      "VRAM used: 28.59 GB\n",
+      "VRAM used: 15.08 GB\n",
       "\n",
       "Loaded 50861 rows\n",
       "Running in TEST MODE with 10 samples\n",
@@ -1867,7 +1891,151 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "Annotating:   0%|          | 0/10 [00:00<?, ?it/s]The following generation flags are not valid and may be ignored: ['early_stopping']. Set `TRANSFORMERS_VERBOSITY=info` for more details.\n"
+      "Annotating:   0%|          | 0/10 [00:00<?, ?it/s]The following generation flags are not valid and may be ignored: ['early_stopping']. Set `TRANSFORMERS_VERBOSITY=info` for more details.\n",
+      "Annotating:  20%|██        | 2/10 [00:03<00:13,  1.66s/it]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "📝 BLOOMZ Debug #1:\n",
+      "Prompt: Instruction: Extract key data on a person based on the name and hints.\n",
+      "You must respond with exactly 5 numbered lines in this format:\n",
+      "1. Full legal name\n",
+      "2. Stage names/aliases  \n",
+      "3. Gender\n",
+      "4. Professio...\n",
+      "Full output: Instruction: Extract key data on a person based on the name and hints.\n",
+      "You must respond with exactly 5 numbered lines in this format:\n",
+      "1. Full legal name\n",
+      "2. Stage names/aliases  \n",
+      "3. Gender\n",
+      "4. Professions (comma-separated, choose ONLY from: actor, adult performer, singer/musician, model, online personality, public figure, voice actor/ASMR, sports professional, tv personality)\n",
+      "5. Country\n",
+      "\n",
+      "Given 'IU' (celebrity, girl, photorealistic, female, asian), provide:\n",
+      "1. Full legal name (Western order if non-...\n",
+      "Generated text:  \n",
+      " IU, celebrity, girl, photorealistic, female, asian\n",
+      "============================================================\n",
+      "\n",
+      "\n",
+      "📝 BLOOMZ Debug #2:\n",
+      "Prompt: Instruction: Extract key data on a person based on the name and hints.\n",
+      "You must respond with exactly 5 numbered lines in this format:\n",
+      "1. Full legal name\n",
+      "2. Stage names/aliases  \n",
+      "3. Gender\n",
+      "4. Professio...\n",
+      "Full output: Instruction: Extract key data on a person based on the name and hints.\n",
+      "You must respond with exactly 5 numbered lines in this format:\n",
+      "1. Full legal name\n",
+      "2. Stage names/aliases  \n",
+      "3. Gender\n",
+      "4. Professions (comma-separated, choose ONLY from: actor, adult performer, singer/musician, model, online personality, public figure, voice actor/ASMR, sports professional, tv personality)\n",
+      "5. Country\n",
+      "\n",
+      "Given 'Super Pose Book' (sexy, female, general purpose, poses, art style), provide:\n",
+      "1. Full legal name (Western...\n",
+      "Generated text:  \n",
+      "Alice Smith\n",
+      "============================================================\n",
+      "\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Annotating:  30%|███       | 3/10 [00:04<00:08,  1.19s/it]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\n",
+      "📝 BLOOMZ Debug #3:\n",
+      "Prompt: Instruction: Extract key data on a person based on the name and hints.\n",
+      "You must respond with exactly 5 numbered lines in this format:\n",
+      "1. Full legal name\n",
+      "2. Stage names/aliases  \n",
+      "3. Gender\n",
+      "4. Professio...\n",
+      "Full output: Instruction: Extract key data on a person based on the name and hints.\n",
+      "You must respond with exactly 5 numbered lines in this format:\n",
+      "1. Full legal name\n",
+      "2. Stage names/aliases  \n",
+      "3. Gender\n",
+      "4. Professions (comma-separated, choose ONLY from: actor, adult performer, singer/musician, model, online personality, public figure, voice actor/ASMR, sports professional, tv personality)\n",
+      "5. Country\n",
+      "\n",
+      "Given 'Liyuu' (girl, photorealistic, portraits, real person), provide:\n",
+      "1. Full legal name (Western order if non...\n",
+      "Generated text:  \n",
+      "Liyuu is a girl who has been a model, singer/musician, and actor. She is also known as \"Lilith\"\n",
+      "============================================================\n",
+      "\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Annotating: 100%|██████████| 10/10 [00:06<00:00,  1.64it/s]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "✅ Progress saved after 10 rows\n",
+      "\n",
+      "✅ Done! Final results saved to /shares/weddigen.ki.uzh/laura_wagner/phase_01/pm-paper/data/CSV/bloomz_annotated_POI_test.csv\n",
+      "\n",
+      "=== Summary Statistics ===\n",
+      "Total processed: 10\n",
+      "\n",
+      "Gender distribution:\n",
+      "gender\n",
+      "Unknown    10\n",
+      "Name: count, dtype: int64\n",
+      "\n",
+      "Top 10 profession combinations:\n",
+      "profession_llm\n",
+      "Unknown    10\n",
+      "Name: count, dtype: int64\n",
+      "\n",
+      "Top 10 countries:\n",
+      "country\n",
+      "Unknown    10\n",
+      "Name: count, dtype: int64\n",
+      "\n",
+      "=== Sample Results ===\n",
+      "      real_name full_name  gender profession_llm country\n",
+      "             IU   Unknown Unknown        Unknown Unknown\n",
+      "Super Pose Book   Unknown Unknown        Unknown Unknown\n",
+      "          Liyuu   Unknown Unknown        Unknown Unknown\n",
+      "          Irene   Unknown Unknown        Unknown Unknown\n",
+      "   AESPA Karina   Unknown Unknown        Unknown Unknown\n",
+      " Saika Kawakita   Unknown Unknown        Unknown Unknown\n",
+      "      Liu Yifei   Unknown Unknown        Unknown Unknown\n",
+      " HashimotoKanna   Unknown Unknown        Unknown Unknown\n",
+      "    Emma Watson   Unknown Unknown        Unknown Unknown\n",
+      "      Gal Gadot   Unknown Unknown        Unknown Unknown\n",
+      "\n",
+      "Total time: 0.1 minutes\n",
+      "Average speed: 5801.5 samples/hour\n",
+      "Final VRAM usage: 15.08 GB\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n"
      ]
     }
    ],

misc/deepseek_query_index.txt

@@ -1 +0,0 @@
-10

misc/gemma_local_query_index.txt

@@ -1 +0,0 @@
-10

misc/mistral_local_query_index.txt

@@ -1 +0,0 @@
-10

misc/qwen_local_query_index.txt

@@ -1 +0,0 @@
-10

misc/qwen_query_index.txt

@@ -1 +0,0 @@
-10