notebooks/run_inference_and_eval.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "2cfb571e-29d9-4115-84d3-560fd499f907",
   "metadata": {},
   "source": [
    "# About this notebook\n",
    "In this notebook, we introduce the overall flow of predicting bandgaps from crystal structures using the fine-tuned LLM (`ysuz/Mistral-Nemo-Base-2407-bandgap`).\n",
    "\n",
    "If you are running this notebook in Google Colab or in your own environment, please place the [train_structure2property.py](/~https://github.com/resnant/llm4mat-tutorial/blob/main/train_structure2property.py) file in the same directory as this notebook, or refer to its contents and copy and paste the necessary functions (load_mp_pickles, get_structure_info_text, prepare_datasets) into the notebook cells."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "7045f72f-e235-4d25-994e-fec043088a6f",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-07-24 10:50:14.886461: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:485] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered\n",
      "2024-07-24 10:50:14.900820: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:8454] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered\n",
      "2024-07-24 10:50:14.905200: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1452] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n",
      "2024-07-24 10:50:14.915864: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
      "To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2024-07-24 10:50:16.270456: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[2024-07-24 10:50:17,634] [INFO] [real_accelerator.py:203:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/compiler_compat/ld: cannot find -laio: No such file or directory\n",
      "collect2: error: ld returned 1 exit status\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[93m [WARNING] \u001b[0m async_io requires the dev libaio .so object and headers but these were not found.\n",
      "\u001b[93m [WARNING] \u001b[0m async_io: please install the libaio-dev package with apt\n",
      "\u001b[93m [WARNING] \u001b[0m If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.\n",
      "\u001b[93m [WARNING] \u001b[0m Please specify the CUTLASS repo directory as environment variable $CUTLASS_PATH\n",
      "\u001b[93m [WARNING] \u001b[0m sparse_attn requires a torch version >= 1.5 and < 2.0 but detected 2.2\n",
      "\u001b[93m [WARNING] \u001b[0m using untested triton version (2.2.0), only 1.0.0 is known to be compatible\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import sys\n",
    "sys.path.append(\"/workspace\")\n",
    "\n",
    "import json\n",
    "import glob\n",
    "import pickle\n",
    "from tqdm import tqdm\n",
    "from multiprocessing import Pool\n",
    "import pandas as pd\n",
    "from transformers import AutoModelForCausalLM, AutoTokenizer\n",
    "from datasets import Dataset\n",
    "import torch\n",
    "from sklearn.metrics import mean_absolute_error, r2_score\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from train_structure2property import load_mp_pickles, get_structure_info_text, prepare_datasets\n",
    "\n",
    "def parse_llm_output(output_text):\n",
    "    try:\n",
    "        result_val = float(output_text.split(\"Output:\\n\")[-1])\n",
    "    except:\n",
    "        result_val= None\n",
    "    return result_val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6b90788b-bd35-4246-9823-10a871a18a0c",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "mp_data = load_mp_pickles(\"../mp_download\")    # tokenizer, model = setup_tokenizer_and_model(model_id = \"mistralai/Mixtral-8x7B-v0.1\")\n",
    "dataset_split = prepare_datasets(mp_data, \"band_gap\")\n",
    "\n",
    "model_id = \"ysuz/Mistral-Nemo-Base-2407-bandgap\"\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_id)\n",
    "model = AutoModelForCausalLM.from_pretrained(model_id,\n",
    "                                             device_map=\"auto\",\n",
    "                                             torch_dtype=torch.float16,\n",
    "                                             # load_in_8bit=True # please use this option instead of torch_dtype=torch.float16 to reduce memory usage\n",
    "                                         )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "557e8599-ce6d-49ed-8dd3-08a2df4aa99d",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Setting `pad_token_id` to `eos_token_id`:2 for open-end generation.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generated raw text:\n",
      "Instruction: What is the bandgap value of following material?:\n",
      "\n",
      "Reduced Formula: BaSrI4\n",
      "abc   :   5.807091   5.807091   8.251028\n",
      "angles:  90.000000  90.000000  90.000000\n",
      "pbc   :       True       True       True\n",
      "space group: ('P4/mmm', 123)\n",
      "Sites (6)\n",
      "  #  SP      a    b         c    magmom\n",
      "  0  Ba    0.5  0.5  0               -0\n",
      "  1  Sr    0    0    0.5             -0\n",
      "  2  I     0    0.5  0.257945         0\n",
      "  3  I     0.5  0    0.257945         0\n",
      "  4  I     0    0.5  0.742055         0\n",
      "  5  I     0.5  0    0.742055         0\n",
      "\n",
      "Output:\n",
      "\n",
      "\n",
      "Output:\n",
      "3.4937\n",
      "\n",
      "\n",
      "- - - - - - - - - - - - -\n",
      "Parsed result: 3.4937 eV\n"
     ]
    }
   ],
   "source": [
    "# example of input context and generation result\n",
    "\n",
    "structure_text = \"\"\"\n",
    "Reduced Formula: BaSrI4\n",
    "abc   :   5.807091   5.807091   8.251028\n",
    "angles:  90.000000  90.000000  90.000000\n",
    "pbc   :       True       True       True\n",
    "space group: ('P4/mmm', 123)\n",
    "Sites (6)\n",
    "  #  SP      a    b         c    magmom\n",
    "  0  Ba    0.5  0.5  0               -0\n",
    "  1  Sr    0    0    0.5             -0\n",
    "  2  I     0    0.5  0.257945         0\n",
    "  3  I     0.5  0    0.257945         0\n",
    "  4  I     0    0.5  0.742055         0\n",
    "  5  I     0.5  0    0.742055         0\n",
    "\n",
    "Output:\n",
    "\"\"\"\n",
    "\n",
    "prompt = f\"Instruction: What is the bandgap value of following material?:\\n{structure_text}\\n\\nOutput:\\n\"\n",
    "inputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\n",
    "\n",
    "with torch.no_grad():\n",
    "    tokens = model.generate(\n",
    "        **inputs,\n",
    "        max_new_tokens=256,\n",
    "        do_sample=True,\n",
    "        temperature=0.5,\n",
    "        top_p=0.9,\n",
    "        repetition_penalty=1.05,\n",
    "    )\n",
    "generated_text = tokenizer.decode(tokens[0], skip_special_tokens=True)\n",
    "print(f\"Generated raw text:\\n{generated_text}\\n\\n\")\n",
    "\n",
    "parsed_text = parse_llm_output(generated_text)\n",
    "print(\"- - - - - - - - - - - - -\")\n",
    "print(f\"Parsed result: {parsed_text} eV\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "f014ab31-ebb8-44ad-ae4a-d9acce1c8a05",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Instruction: What is the bandgap value of following material?:\n",
      "Reduced Formula: BaSrI4\n",
      "abc   :   5.807091   5.807091   8.251028\n",
      "angles:  90.000000  90.000000  90.000000\n",
      "pbc   :       True       True       True\n",
      "space group: ('P4/mmm', 123)\n",
      "Sites (6)\n",
      "  #  SP      a    b         c    magmom\n",
      "  0  Ba    0.5  0.5  0               -0\n",
      "  1  Sr    0    0    0.5             -0\n",
      "  2  I     0    0.5  0.257945         0\n",
      "  3  I     0.5  0    0.257945         0\n",
      "  4  I     0    0.5  0.742055         0\n",
      "  5  I     0.5  0    0.742055         0\n",
      "\n",
      "Output:\n",
      "3.2021999999999995\n"
     ]
    }
   ],
   "source": [
    "# example of input context\n",
    "i = 4\n",
    "structure_text = dataset_split[\"test\"][i][\"structure\"]\n",
    "target = dataset_split[\"test\"][i][\"target\"]\n",
    "prompt = f\"Instruction: What is the bandgap value of following material?:\\n{structure_text}\\n\\nOutput:\\n{target}\"\n",
    "\n",
    "print(prompt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a39142d6-d51e-4547-afe1-bedc34566545",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "127it [00:27,  4.57it/s]\n"
     ]
    }
   ],
   "source": [
    "# batch inference on test dataset\n",
    "\n",
    "generated_data = [] \n",
    "\n",
    "for mat_idx, structure_text in tqdm(enumerate([material[\"structure\"] for material in dataset_split[\"test\"]][::100])): # sample 1% of dataset\n",
    "    prompt = f\"Instruction: What is the bandgap value of following material?:\\n{structure_text}\\n\\nOutput:\\n\"\n",
    "    \n",
    "    inputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\n",
    "    with torch.no_grad():\n",
    "        tokens = model.generate(\n",
    "            **inputs,\n",
    "            max_new_tokens=256,\n",
    "            do_sample=True,\n",
    "            temperature=0.5,\n",
    "            top_p=0.9,\n",
    "            repetition_penalty=1.05,\n",
    "            pad_token_id=tokenizer.pad_token_id,\n",
    "        )\n",
    "    generated_text = tokenizer.decode(tokens[0], skip_special_tokens=True)\n",
    "    parsed_text = parse_llm_output(generated_text)\n",
    "    \n",
    "    generated_data.append({\"mat_idx\": mat_idx, \"structure_text\": structure_text, \"output\": generated_text, \"parsed_text\":parsed_text})\n",
    "\n",
    "pred_values = [dt[\"parsed_text\"] for dt in generated_data]\n",
    "true_values = [material[\"target\"] for material in dataset_split[\"test\"]][::100] # sample 1% of dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "41504055-37c0-4cab-9525-f324a716aec2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "R2 Score: 0.7923199390551515\n",
      "Mean Absolute Error: 0.3578055118110236 eV\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x800 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "r2 = r2_score(true_values, pred_values)\n",
    "mae = mean_absolute_error(true_values, pred_values)\n",
    "print(f'R2 Score: {r2}')\n",
    "print(f'Mean Absolute Error: {mae} eV')\n",
    "\n",
    "plt.figure(figsize=(8, 8))\n",
    "plt.scatter(true_values, pred_values)\n",
    "plt.plot([min(true_values), max(true_values)], [min(true_values), max(true_values)], 'k--', lw=1)\n",
    "plt.xlabel('True Values (eV)')\n",
    "plt.ylabel('Predicted Values (eV)')\n",
    "plt.title('y-y plot of bandgap')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "66f75fec-81b3-40cd-90bf-e404ad257104",
   "metadata": {},
   "source": [
    "## Note\n",
    "When considering the current top-tier DNN models, which are primarily based on graph neural networks, they achieve an approximate MAE of 0.2 eV for bandgap prediction (Table 1). Therefore, our performance (MAE = 0.36 eV) is promising as an initial result with large language models.\n",
    "\n",
    "This table is taken from our recent work, [Crystalformer: Infinitely Connected Attention for Periodic Structure Encoding](https://omron-sinicx.github.io/crystalformer/)\n",
    "\n",
    "![materials property prediction benchmarking](https://omron-sinicx.github.io/crystalformer/assets/benchmark.png)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "05361b09-6920-436e-a815-090fdf2d2ade",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}