{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from functools import lru_cache\n",
    "from concurrent.futures import ThreadPoolExecutor\n",
    "import promptquality as pq\n",
    "from dotenv import load_dotenv\n",
    "from data_loader import DATASETS, load_data\n",
    "from tqdm.auto import tqdm\n",
    "\n",
    "load_dotenv()\n",
    "pq.login(\"https://console.demo.rungalileo.io\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "project_name = \"agent-lb-v1\"\n",
    "PROJECT_ID = pq.get_project_from_name(project_name).id\n",
    "\n",
    "\n",
    "@lru_cache(maxsize=1000)\n",
    "def get_output_df(model, dataset):\n",
    "    print(f\"Getting metrics for {model} {project_name} for dataset {dataset}\")\n",
    "    run_name = f\"{model} {dataset}\"\n",
    "    run_id = pq.get_run_from_name(run_name, PROJECT_ID).id\n",
    "    rows = pq.get_rows(\n",
    "        project_id=PROJECT_ID,\n",
    "        run_id=run_id,\n",
    "        task_type=None,\n",
    "        config=None,\n",
    "        starting_token=0,\n",
    "        limit=1000,\n",
    "    )\n",
    "\n",
    "    rationales = [d.metrics.tool_selection_quality_rationale for d in rows]\n",
    "\n",
    "    scores = [\n",
    "        round(d.metrics.tool_selection_quality, 2)\n",
    "        for d, rationale in zip(rows, rationales)\n",
    "        if rationale\n",
    "    ]\n",
    "    \n",
    "    explanations = [\n",
    "        d.metrics.tool_selection_quality_explanation\n",
    "        for d, rationale in zip(rows, rationales)\n",
    "        if rationale\n",
    "    ]\n",
    "    \n",
    "    responses = [d.response for d, rationale in zip(rows, rationales)\n",
    "        if rationale\n",
    "    ]\n",
    "    \n",
    "    rationales = [r for r in rationales if r]\n",
    "    mean_score = round(np.mean(scores), 2)\n",
    "    \n",
    "    data = {\n",
    "        \"response\": responses,\n",
    "        \"mean_score\": mean_score,\n",
    "        \"score\": scores,\n",
    "        \"rationale\": rationales,\n",
    "        \"explanation\": explanations,\n",
    "    }\n",
    "    return pd.DataFrame(data)\n",
    "\n",
    "def save_output_df(df, model, dataset):\n",
    "    os.makedirs(f\"output/{model}\", exist_ok=True)\n",
    "    df.to_parquet(f\"output/{model}/{dataset}.parquet\")\n",
    "\n",
    "def get_updated_df(df, df_output):\n",
    "    df = df.iloc[:len(df_output)].copy()\n",
    "    \n",
    "    df[\"response\"] = df_output[\"response\"].tolist()\n",
    "    df[\"rationale\"] = df_output[\"rationale\"].tolist()\n",
    "    df[\"explanation\"] = df_output[\"explanation\"].tolist()\n",
    "    df[\"score\"] = df_output[\"score\"].tolist()\n",
    "    cols = ['conversation', 'tools_langchain', 'n_turns',\n",
    "            'len_query', 'n_tools', 'response', 'rationale', 'explanation', 'score']\n",
    "    return df[cols]\n",
    "\n",
    "\n",
    "def get_chat_and_score_df(model, dataset):\n",
    "    df_output = pd.read_parquet(f\"output/{model}/{dataset}.parquet\")\n",
    "    df = pd.read_parquet(f\"datasets/{dataset}.parquet\")\n",
    "    df = get_updated_df(df, df_output)\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def process_dataset(args):\n",
    "    model, dataset = args\n",
    "    if os.path.exists(f\"output/{model}/{dataset}.parquet\"):\n",
    "        return None\n",
    "    print(model, dataset)\n",
    "    df_output = get_output_df(model, dataset)\n",
    "    save_output_df(df_output, model, dataset)\n",
    "    return f\"Completed: {model} - {dataset}\"\n",
    "\n",
    "def process_model_datasets(model, datasets, max_workers=5):\n",
    "    with ThreadPoolExecutor(max_workers=max_workers) as executor:\n",
    "        # Create arguments list for each dataset\n",
    "        args_list = [(model, dataset) for dataset in datasets]\n",
    "        \n",
    "        # Process datasets in parallel with progress bar\n",
    "        list(tqdm(\n",
    "            executor.map(process_dataset, args_list),\n",
    "            total=len(datasets),\n",
    "            desc=f\"Datasets ({model})\",\n",
    "            position=1,\n",
    "            leave=False\n",
    "        ))\n",
    "\n",
    "\n",
    "models = [\"accounts/fireworks/models/qwen2p5-72b-instruct\", \"meta-llama/Llama-3.3-70B-Instruct-Turbo\", \"meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo\"]\n",
    "# models = load_data()[\"Model\"]\n",
    "\n",
    "# Process each model sequentially, but datasets in parallel\n",
    "for model in tqdm(models, desc=\"Models\", position=0):\n",
    "    process_model_datasets(model, DATASETS)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "langgraph",
   "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": "3.12.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}