{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e119e71a-f88a-4d5c-90fb-e60b84c4f42c",
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import (\n",
    "    AutoModelForCausalLM,\n",
    "    AutoTokenizer,\n",
    "    AutoTokenizer,\n",
    ")\n",
    "from peft import PeftModel, PeftConfig\n",
    "import torch\n",
    "\n",
    "d_map = {\"\": torch.cuda.current_device()} if torch.cuda.is_available() else None\n",
    "local_model_path = \"outputs/checkpoint-100\"     # Path to the combined weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "ba591ab9-5029-46e8-b9a9-428de3896e62",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "ee8258a57258444baf79b52af6444788",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "ename": "OutOfMemoryError",
     "evalue": "CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacty of 23.68 GiB of which 79.62 MiB is free. Process 657358 has 23.59 GiB memory in use. Of the allocated memory 23.00 GiB is allocated by PyTorch, and 357.02 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mOutOfMemoryError\u001b[0m                          Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[23], line 4\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;66;03m# Loading the base Model\u001b[39;00m\n\u001b[1;32m      2\u001b[0m config \u001b[38;5;241m=\u001b[39m PeftConfig\u001b[38;5;241m.\u001b[39mfrom_pretrained(local_model_path)\n\u001b[0;32m----> 4\u001b[0m model \u001b[38;5;241m=\u001b[39m \u001b[43mAutoModelForCausalLM\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfrom_pretrained\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m      5\u001b[0m \u001b[43m    \u001b[49m\u001b[43mconfig\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbase_model_name_or_path\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\n\u001b[1;32m      6\u001b[0m \u001b[43m    \u001b[49m\u001b[43mreturn_dict\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m,\u001b[49m\n\u001b[1;32m      7\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtorch_dtype\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtorch\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfloat16\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m      8\u001b[0m \u001b[43m    \u001b[49m\u001b[38;5;66;43;03m# load_in_4bit=True, \u001b[39;49;00m\n\u001b[1;32m      9\u001b[0m \u001b[43m    \u001b[49m\u001b[43mdevice_map\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43md_map\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m     10\u001b[0m \u001b[43m)\u001b[49m\n\u001b[1;32m     11\u001b[0m tokenizer \u001b[38;5;241m=\u001b[39m AutoTokenizer\u001b[38;5;241m.\u001b[39mfrom_pretrained(config\u001b[38;5;241m.\u001b[39mbase_model_name_or_path)\n",
      "File \u001b[0;32m/usr/local/lib/python3.10/dist-packages/transformers/models/auto/auto_factory.py:566\u001b[0m, in \u001b[0;36m_BaseAutoModelClass.from_pretrained\u001b[0;34m(cls, pretrained_model_name_or_path, *model_args, **kwargs)\u001b[0m\n\u001b[1;32m    564\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m \u001b[38;5;28mtype\u001b[39m(config) \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mcls\u001b[39m\u001b[38;5;241m.\u001b[39m_model_mapping\u001b[38;5;241m.\u001b[39mkeys():\n\u001b[1;32m    565\u001b[0m     model_class \u001b[38;5;241m=\u001b[39m _get_model_class(config, \u001b[38;5;28mcls\u001b[39m\u001b[38;5;241m.\u001b[39m_model_mapping)\n\u001b[0;32m--> 566\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mmodel_class\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfrom_pretrained\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    567\u001b[0m \u001b[43m        \u001b[49m\u001b[43mpretrained_model_name_or_path\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mmodel_args\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mconfig\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mconfig\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mhub_kwargs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\n\u001b[1;32m    568\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    569\u001b[0m \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mValueError\u001b[39;00m(\n\u001b[1;32m    570\u001b[0m     \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mUnrecognized configuration class \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mconfig\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m for this kind of AutoModel: \u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mcls\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m    571\u001b[0m     \u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mModel type should be one of \u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;124m'\u001b[39m\u001b[38;5;124m, \u001b[39m\u001b[38;5;124m'\u001b[39m\u001b[38;5;241m.\u001b[39mjoin(c\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mfor\u001b[39;00m\u001b[38;5;250m \u001b[39mc\u001b[38;5;250m \u001b[39m\u001b[38;5;129;01min\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28mcls\u001b[39m\u001b[38;5;241m.\u001b[39m_model_mapping\u001b[38;5;241m.\u001b[39mkeys())\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.\u001b[39m\u001b[38;5;124m\"\u001b[39m\n\u001b[1;32m    572\u001b[0m )\n",
      "File \u001b[0;32m/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py:3850\u001b[0m, in \u001b[0;36mPreTrainedModel.from_pretrained\u001b[0;34m(cls, pretrained_model_name_or_path, config, cache_dir, ignore_mismatched_sizes, force_download, local_files_only, token, revision, use_safetensors, *model_args, **kwargs)\u001b[0m\n\u001b[1;32m   3841\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m dtype_orig \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m   3842\u001b[0m         torch\u001b[38;5;241m.\u001b[39mset_default_dtype(dtype_orig)\n\u001b[1;32m   3843\u001b[0m     (\n\u001b[1;32m   3844\u001b[0m         model,\n\u001b[1;32m   3845\u001b[0m         missing_keys,\n\u001b[1;32m   3846\u001b[0m         unexpected_keys,\n\u001b[1;32m   3847\u001b[0m         mismatched_keys,\n\u001b[1;32m   3848\u001b[0m         offload_index,\n\u001b[1;32m   3849\u001b[0m         error_msgs,\n\u001b[0;32m-> 3850\u001b[0m     ) \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mcls\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_load_pretrained_model\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m   3851\u001b[0m \u001b[43m        \u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3852\u001b[0m \u001b[43m        \u001b[49m\u001b[43mstate_dict\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3853\u001b[0m \u001b[43m        \u001b[49m\u001b[43mloaded_state_dict_keys\u001b[49m\u001b[43m,\u001b[49m\u001b[43m  \u001b[49m\u001b[38;5;66;43;03m# XXX: rename?\u001b[39;49;00m\n\u001b[1;32m   3854\u001b[0m \u001b[43m        \u001b[49m\u001b[43mresolved_archive_file\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3855\u001b[0m \u001b[43m        \u001b[49m\u001b[43mpretrained_model_name_or_path\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3856\u001b[0m \u001b[43m        \u001b[49m\u001b[43mignore_mismatched_sizes\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mignore_mismatched_sizes\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3857\u001b[0m \u001b[43m        \u001b[49m\u001b[43msharded_metadata\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43msharded_metadata\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3858\u001b[0m \u001b[43m        \u001b[49m\u001b[43m_fast_init\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m_fast_init\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3859\u001b[0m \u001b[43m        \u001b[49m\u001b[43mlow_cpu_mem_usage\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mlow_cpu_mem_usage\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3860\u001b[0m \u001b[43m        \u001b[49m\u001b[43mdevice_map\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdevice_map\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3861\u001b[0m \u001b[43m        \u001b[49m\u001b[43moffload_folder\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moffload_folder\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3862\u001b[0m \u001b[43m        \u001b[49m\u001b[43moffload_state_dict\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moffload_state_dict\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3863\u001b[0m \u001b[43m        \u001b[49m\u001b[43mdtype\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtorch_dtype\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3864\u001b[0m \u001b[43m        \u001b[49m\u001b[43mis_quantized\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mgetattr\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mquantization_method\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43;01mNone\u001b[39;49;00m\u001b[43m)\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m==\u001b[39;49m\u001b[43m \u001b[49m\u001b[43mQuantizationMethod\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mBITS_AND_BYTES\u001b[49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3865\u001b[0m \u001b[43m        \u001b[49m\u001b[43mkeep_in_fp32_modules\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mkeep_in_fp32_modules\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   3866\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   3868\u001b[0m model\u001b[38;5;241m.\u001b[39mis_loaded_in_4bit \u001b[38;5;241m=\u001b[39m load_in_4bit\n\u001b[1;32m   3869\u001b[0m model\u001b[38;5;241m.\u001b[39mis_loaded_in_8bit \u001b[38;5;241m=\u001b[39m load_in_8bit\n",
      "File \u001b[0;32m/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py:4284\u001b[0m, in \u001b[0;36mPreTrainedModel._load_pretrained_model\u001b[0;34m(cls, model, state_dict, loaded_keys, resolved_archive_file, pretrained_model_name_or_path, ignore_mismatched_sizes, sharded_metadata, _fast_init, low_cpu_mem_usage, device_map, offload_folder, offload_state_dict, dtype, is_quantized, keep_in_fp32_modules)\u001b[0m\n\u001b[1;32m   4280\u001b[0m                     set_module_quantized_tensor_to_device(\n\u001b[1;32m   4281\u001b[0m                         model_to_load, key, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mcpu\u001b[39m\u001b[38;5;124m\"\u001b[39m, torch\u001b[38;5;241m.\u001b[39mempty(\u001b[38;5;241m*\u001b[39mparam\u001b[38;5;241m.\u001b[39msize(), dtype\u001b[38;5;241m=\u001b[39mdtype)\n\u001b[1;32m   4282\u001b[0m                     )\n\u001b[1;32m   4283\u001b[0m     \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 4284\u001b[0m         new_error_msgs, offload_index, state_dict_index \u001b[38;5;241m=\u001b[39m \u001b[43m_load_state_dict_into_meta_model\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m   4285\u001b[0m \u001b[43m            \u001b[49m\u001b[43mmodel_to_load\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4286\u001b[0m \u001b[43m            \u001b[49m\u001b[43mstate_dict\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4287\u001b[0m \u001b[43m            \u001b[49m\u001b[43mloaded_keys\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4288\u001b[0m \u001b[43m            \u001b[49m\u001b[43mstart_prefix\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4289\u001b[0m \u001b[43m            \u001b[49m\u001b[43mexpected_keys\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4290\u001b[0m \u001b[43m            \u001b[49m\u001b[43mdevice_map\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdevice_map\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4291\u001b[0m \u001b[43m            \u001b[49m\u001b[43moffload_folder\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moffload_folder\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4292\u001b[0m \u001b[43m            \u001b[49m\u001b[43moffload_index\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moffload_index\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4293\u001b[0m \u001b[43m            \u001b[49m\u001b[43mstate_dict_folder\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mstate_dict_folder\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4294\u001b[0m \u001b[43m            \u001b[49m\u001b[43mstate_dict_index\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mstate_dict_index\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4295\u001b[0m \u001b[43m            \u001b[49m\u001b[43mdtype\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mdtype\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4296\u001b[0m \u001b[43m            \u001b[49m\u001b[43mis_quantized\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mis_quantized\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4297\u001b[0m \u001b[43m            \u001b[49m\u001b[43mis_safetensors\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mis_safetensors\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4298\u001b[0m \u001b[43m            \u001b[49m\u001b[43mkeep_in_fp32_modules\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mkeep_in_fp32_modules\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4299\u001b[0m \u001b[43m            \u001b[49m\u001b[43munexpected_keys\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43munexpected_keys\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m   4300\u001b[0m \u001b[43m        \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   4301\u001b[0m         error_msgs \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m new_error_msgs\n\u001b[1;32m   4302\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n",
      "File \u001b[0;32m/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py:805\u001b[0m, in \u001b[0;36m_load_state_dict_into_meta_model\u001b[0;34m(model, state_dict, loaded_state_dict_keys, start_prefix, expected_keys, device_map, offload_folder, offload_index, state_dict_folder, state_dict_index, dtype, is_quantized, is_safetensors, keep_in_fp32_modules, unexpected_keys)\u001b[0m\n\u001b[1;32m    802\u001b[0m     state_dict_index \u001b[38;5;241m=\u001b[39m offload_weight(param, param_name, state_dict_folder, state_dict_index)\n\u001b[1;32m    803\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m is_quantized:\n\u001b[1;32m    804\u001b[0m     \u001b[38;5;66;03m# For backward compatibility with older versions of `accelerate`\u001b[39;00m\n\u001b[0;32m--> 805\u001b[0m     \u001b[43mset_module_tensor_to_device\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mparam_name\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mparam_device\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mset_module_kwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    806\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m param\u001b[38;5;241m.\u001b[39mdtype \u001b[38;5;129;01min\u001b[39;00m (torch\u001b[38;5;241m.\u001b[39mint8, torch\u001b[38;5;241m.\u001b[39muint8) \u001b[38;5;129;01mand\u001b[39;00m is_quantized:\n\u001b[1;32m    807\u001b[0m     \u001b[38;5;66;03m# handling newly quantized weights and loaded quantized weights\u001b[39;00m\n\u001b[1;32m    808\u001b[0m     \u001b[38;5;66;03m# edit the param.dtype restrictions and is_quantized condition when adding new quant methods\u001b[39;00m\n\u001b[1;32m    809\u001b[0m     quantized_stats \u001b[38;5;241m=\u001b[39m {}\n",
      "File \u001b[0;32m/usr/local/lib/python3.10/dist-packages/accelerate/utils/modeling.py:347\u001b[0m, in \u001b[0;36mset_module_tensor_to_device\u001b[0;34m(module, tensor_name, device, value, dtype, fp16_statistics)\u001b[0m\n\u001b[1;32m    345\u001b[0m             module\u001b[38;5;241m.\u001b[39m_parameters[tensor_name] \u001b[38;5;241m=\u001b[39m param_cls(new_value, requires_grad\u001b[38;5;241m=\u001b[39mold_value\u001b[38;5;241m.\u001b[39mrequires_grad)\n\u001b[1;32m    346\u001b[0m \u001b[38;5;28;01melif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(value, torch\u001b[38;5;241m.\u001b[39mTensor):\n\u001b[0;32m--> 347\u001b[0m     new_value \u001b[38;5;241m=\u001b[39m \u001b[43mvalue\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mto\u001b[49m\u001b[43m(\u001b[49m\u001b[43mdevice\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    348\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m    349\u001b[0m     new_value \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mtensor(value, device\u001b[38;5;241m=\u001b[39mdevice)\n",
      "\u001b[0;31mOutOfMemoryError\u001b[0m: CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacty of 23.68 GiB of which 79.62 MiB is free. Process 657358 has 23.59 GiB memory in use. Of the allocated memory 23.00 GiB is allocated by PyTorch, and 357.02 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF"
     ]
    }
   ],
   "source": [
    "# Loading the base Model\n",
    "config = PeftConfig.from_pretrained(local_model_path)\n",
    "\n",
    "model = AutoModelForCausalLM.from_pretrained(\n",
    "    config.base_model_name_or_path, \n",
    "    return_dict=True,\n",
    "    torch_dtype=torch.float16,\n",
    "    # load_in_4bit=True, \n",
    "    device_map=d_map,\n",
    ")\n",
    "tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "c1d36c14-0bfc-4215-8576-bb390a3a6114",
   "metadata": {},
   "outputs": [],
   "source": [
    "# load the base model with the Lora model\n",
    "model = PeftModel.from_pretrained(model, local_model_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "72cdeaa8-4b0c-45d0-a585-2f86626e280b",
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "09fa4575-0dec-4e62-a43f-77e57f68c4a9",
   "metadata": {},
   "outputs": [],
   "source": [
    "def inferance(query: str, model, tokenizer, temp = 1.0, limit = 200) -> str:\n",
    "  device = d_map\n",
    "\n",
    "  prompt_template = \"\"\"\n",
    "  Below is an instruction that describes a task. Write a response that appropriately completes the request.\n",
    "  ### Question:\n",
    "  {query}\n",
    "\n",
    "  ### Answer:\n",
    "  \"\"\"\n",
    "  prompt = prompt_template.format(query=query)\n",
    "\n",
    "  encodeds = tokenizer(prompt, return_tensors=\"pt\", add_special_tokens=True)\n",
    "\n",
    "  model_inputs = encodeds.to(device)\n",
    "\n",
    "  generated_ids = model.generate(**model_inputs, max_new_tokens=int(limit), temperature=temp, do_sample=True, pad_token_id=tokenizer.eos_token_id)\n",
    "  decoded = tokenizer.batch_decode(generated_ids)\n",
    "  return (decoded[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ba47700b-0787-4677-a5a1-c1a1b4063fe2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running on local URL:  http://127.0.0.1:7862\n",
      "Running on public URL: https://f4d39c90e01dcf849c.gradio.live\n",
      "\n",
      "This share link expires in 72 hours. For free permanent hosting and GPU upgrades, run `gradio deploy` from Terminal to deploy to Spaces (https://huggingface.co/spaces)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div><iframe src=\"https://f4d39c90e01dcf849c.gradio.live\" width=\"100%\" height=\"500\" allow=\"autoplay; camera; microphone; clipboard-read; clipboard-write;\" frameborder=\"0\" allowfullscreen></iframe></div>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": []
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Attempting to cast a BatchEncoding to type {'': 0}. This is not supported.\n",
      "/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py:1413: UserWarning: You are calling .generate() with the `input_ids` being on a device type different than your model's device. `input_ids` is on cpu, whereas the model is on cuda. You may experience unexpected behaviors or slower generation. Please make sure that you have put `input_ids` to the correct device by calling for example input_ids = input_ids.to('cuda') before running `.generate()`.\n",
      "  warnings.warn(\n",
      "Traceback (most recent call last):\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/gradio/queueing.py\", line 495, in call_prediction\n",
      "    output = await route_utils.call_process_api(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/gradio/route_utils.py\", line 230, in call_process_api\n",
      "    output = await app.get_blocks().process_api(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/gradio/blocks.py\", line 1590, in process_api\n",
      "    result = await self.call_function(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/gradio/blocks.py\", line 1176, in call_function\n",
      "    prediction = await anyio.to_thread.run_sync(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/anyio/to_thread.py\", line 33, in run_sync\n",
      "    return await get_async_backend().run_sync_in_worker_thread(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/anyio/_backends/_asyncio.py\", line 2106, in run_sync_in_worker_thread\n",
      "    return await future\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/anyio/_backends/_asyncio.py\", line 833, in run\n",
      "    result = context.run(func, *args)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/gradio/utils.py\", line 678, in wrapper\n",
      "    response = f(*args, **kwargs)\n",
      "  File \"/tmp/ipykernel_812/3792119410.py\", line 5, in predict\n",
      "    out = inferance(prompt, model, tokenizer, temp = 1.0, limit = 200)\n",
      "  File \"/tmp/ipykernel_812/3078938966.py\", line 17, in inferance\n",
      "    generated_ids = model.generate(**model_inputs, max_new_tokens=int(limit), temperature=temp, do_sample=True, pad_token_id=tokenizer.eos_token_id)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/peft/peft_model.py\", line 1140, in generate\n",
      "    outputs = self.base_model.generate(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py\", line 115, in decorate_context\n",
      "    return func(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py\", line 1525, in generate\n",
      "    return self.sample(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py\", line 2622, in sample\n",
      "    outputs = self(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py\", line 1518, in _wrapped_call_impl\n",
      "    return self._call_impl(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py\", line 1527, in _call_impl\n",
      "    return forward_call(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/transformers/models/mistral/modeling_mistral.py\", line 1154, in forward\n",
      "    outputs = self.model(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py\", line 1518, in _wrapped_call_impl\n",
      "    return self._call_impl(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py\", line 1527, in _call_impl\n",
      "    return forward_call(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/transformers/models/mistral/modeling_mistral.py\", line 984, in forward\n",
      "    inputs_embeds = self.embed_tokens(input_ids)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py\", line 1518, in _wrapped_call_impl\n",
      "    return self._call_impl(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py\", line 1527, in _call_impl\n",
      "    return forward_call(*args, **kwargs)\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/modules/sparse.py\", line 162, in forward\n",
      "    return F.embedding(\n",
      "  File \"/usr/local/lib/python3.10/dist-packages/torch/nn/functional.py\", line 2233, in embedding\n",
      "    return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)\n",
      "RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper_CUDA__index_select)\n"
     ]
    }
   ],
   "source": [
    "import gradio as gr\n",
    "\n",
    "def predict(temp, limit, text):\n",
    "    prompt = text\n",
    "    out = inferance(prompt, model, tokenizer, temp = 1.0, limit = 200)\n",
    "    return out\n",
    "\n",
    "pred = gr.Interface(\n",
    "    predict,\n",
    "    inputs=[\n",
    "        gr.Slider(0.001, 10, value=0.1, label=\"Temperature\"),\n",
    "        gr.Slider(1, 1024, value=128, label=\"Token Limit\"),\n",
    "        gr.Textbox(\n",
    "            label=\"Input\",\n",
    "            lines=1,\n",
    "            value=\"#### Human: What's the capital of Australia?#### Assistant: \",\n",
    "        ),\n",
    "    ],\n",
    "    outputs='text',\n",
    ")\n",
    "\n",
    "pred.launch(share=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "62622184-9d02-4bd0-8c4e-d6775ce20f75",
   "metadata": {},
   "outputs": [],
   "source": [
    "###some factors to try\n",
    "from_pt=True"
   ]
  }
 ],
 "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": "3.10.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}