{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100\n",
      "Original Hash: 4adcfc6bb9af0e6eada5d4c212c29b100d58c832a641bff902d931f061e4a32b\n",
      "Modified Hash: c978ccaa0f98f616d249d7ad78f4c5c4534b26e9bc32b6a644e278159bf2a93e\n"
     ]
    }
   ],
   "source": [
    "from safetensors import safe_open\n",
    "from safetensors.torch import save_file\n",
    "import torch\n",
    "import hashlib\n",
    "import random\n",
    "\n",
    "filename = \"model-00003-of-00004.safetensors\"\n",
    "modified_model_name = \"modified_model3.safetensors\"\n",
    "\n",
    "\n",
    "with safe_open(filename, framework=\"pt\") as f:\n",
    "    tensors = {key: f.get_tensor(key) for key in f.keys()}\n",
    "\n",
    "def introduce_noise(tensor, noise_level=1e-8):\n",
    "    noise = torch.randn(tensor.size()) * noise_level\n",
    "    return (tensor + noise).to(tensor.dtype)\n",
    "modified = 0;\n",
    "for key in tensors:\n",
    "    modified +=1\n",
    "    tensors[key] = introduce_noise(tensors[key])\n",
    "print(modified)\n",
    "\n",
    "metadata = {\n",
    "\"format\": \"pt\"  # Adjust based on actual format needed\n",
    "}\n",
    "\n",
    "save_file(tensors, modified_model_name, metadata=metadata)\n",
    "\n",
    "\n",
    "def compute_hash(filename):\n",
    "    hasher = hashlib.sha256()\n",
    "    with open(filename, \"rb\") as f:\n",
    "        buf = f.read()\n",
    "        hasher.update(buf)\n",
    "        return hasher.hexdigest()\n",
    "\n",
    "original_hash = compute_hash(filename)\n",
    "modified_hash = compute_hash(modified_model_name)\n",
    "\n",
    "print(f\"Original Hash: {original_hash}\")\n",
    "print(f\"Modified Hash: {modified_hash}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "104\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[7], line 27\u001b[0m\n\u001b[1;32m     21\u001b[0m \u001b[39mprint\u001b[39m(modified)\n\u001b[1;32m     23\u001b[0m metadata \u001b[39m=\u001b[39m {\n\u001b[1;32m     24\u001b[0m \u001b[39m\"\u001b[39m\u001b[39mformat\u001b[39m\u001b[39m\"\u001b[39m: \u001b[39m\"\u001b[39m\u001b[39mpt\u001b[39m\u001b[39m\"\u001b[39m  \u001b[39m# Adjust based on actual format needed\u001b[39;00m\n\u001b[1;32m     25\u001b[0m }\n\u001b[0;32m---> 27\u001b[0m save_file(tensors, modified_model_name, metadata\u001b[39m=\u001b[39;49mmetadata)\n\u001b[1;32m     30\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39mcompute_hash\u001b[39m(filename):\n\u001b[1;32m     31\u001b[0m     hasher \u001b[39m=\u001b[39m hashlib\u001b[39m.\u001b[39msha256()\n",
      "File \u001b[0;32m/opt/miniconda3/envs/bittensor/lib/python3.10/site-packages/safetensors/torch.py:286\u001b[0m, in \u001b[0;36msave_file\u001b[0;34m(tensors, filename, metadata)\u001b[0m\n\u001b[1;32m    255\u001b[0m \u001b[39mdef\u001b[39;00m \u001b[39msave_file\u001b[39m(\n\u001b[1;32m    256\u001b[0m     tensors: Dict[\u001b[39mstr\u001b[39m, torch\u001b[39m.\u001b[39mTensor],\n\u001b[1;32m    257\u001b[0m     filename: Union[\u001b[39mstr\u001b[39m, os\u001b[39m.\u001b[39mPathLike],\n\u001b[1;32m    258\u001b[0m     metadata: Optional[Dict[\u001b[39mstr\u001b[39m, \u001b[39mstr\u001b[39m]] \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m,\n\u001b[1;32m    259\u001b[0m ):\n\u001b[1;32m    260\u001b[0m \u001b[39m    \u001b[39m\u001b[39m\"\"\"\u001b[39;00m\n\u001b[1;32m    261\u001b[0m \u001b[39m    Saves a dictionary of tensors into raw bytes in safetensors format.\u001b[39;00m\n\u001b[1;32m    262\u001b[0m \n\u001b[0;32m   (...)\u001b[0m\n\u001b[1;32m    284\u001b[0m \u001b[39m    ```\u001b[39;00m\n\u001b[1;32m    285\u001b[0m \u001b[39m    \"\"\"\u001b[39;00m\n\u001b[0;32m--> 286\u001b[0m     serialize_file(_flatten(tensors), filename, metadata\u001b[39m=\u001b[39;49mmetadata)\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "from safetensors import safe_open\n",
    "from safetensors.torch import save_file\n",
    "import torch\n",
    "import hashlib\n",
    "import random\n",
    "\n",
    "filename = \"model-00002-of-00004.safetensors\"\n",
    "modified_model_name = \"modified_model2.safetensors\"\n",
    "\n",
    "\n",
    "with safe_open(filename, framework=\"pt\") as f:\n",
    "    tensors = {key: f.get_tensor(key) for key in f.keys()}\n",
    "\n",
    "def introduce_noise(tensor, noise_level=1e-8):\n",
    "    noise = torch.randn(tensor.size()) * noise_level\n",
    "    return (tensor + noise).to(tensor.dtype)\n",
    "modified = 0;\n",
    "for key in tensors:\n",
    "    modified +=1\n",
    "    tensors[key] = introduce_noise(tensors[key])\n",
    "print(modified)\n",
    "\n",
    "metadata = {\n",
    "\"format\": \"pt\"  # Adjust based on actual format needed\n",
    "}\n",
    "\n",
    "save_file(tensors, modified_model_name, metadata=metadata)\n",
    "\n",
    "\n",
    "def compute_hash(filename):\n",
    "    hasher = hashlib.sha256()\n",
    "    with open(filename, \"rb\") as f:\n",
    "        buf = f.read()\n",
    "        hasher.update(buf)\n",
    "        return hasher.hexdigest()\n",
    "\n",
    "original_hash = compute_hash(filename)\n",
    "modified_hash = compute_hash(modified_model_name)\n",
    "\n",
    "print(f\"Original Hash: {original_hash}\")\n",
    "print(f\"Modified Hash: {modified_hash}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from safetensors import safe_open\n",
    "from safetensors.torch import save_file\n",
    "import hashlib\n",
    "\n",
    "filename = \"model-00002-of-00004.safetensors\"\n",
    "modified_model_name = \"modified_model2.safetensors\"\n",
    "\n",
    "# Load the original model\n",
    "with safe_open(filename, framework=\"pt\") as f:\n",
    "    tensors = {key: f.get_tensor(key) for key in f.keys()}\n",
    "\n",
    "# Define anonymous metadata\n",
    "metadata = {\n",
    "    \"description\": \"This model has anonymous metadata\",  # Neutral description\n",
    "    \"created_by\": \"anonymous\",  # Anonymous creator\n",
    "    \"format\": \"pt\"  # Adjust based on actual format needed\n",
    "}\n",
    "\n",
    "# Save the model with new metadata but without altering weights\n",
    "save_file(tensors, modified_model_name, metadata=metadata)\n",
    "\n",
    "# Function to compute file hash\n",
    "def compute_hash(filename):\n",
    "    hasher = hashlib.sha256()\n",
    "    with open(filename, \"rb\") as f:\n",
    "        buf = f.read()\n",
    "        hasher.update(buf)\n",
    "        return hasher.hexdigest()\n",
    "\n",
    "# Compute hashes\n",
    "original_hash = compute_hash(filename)\n",
    "modified_hash = compute_hash(modified_model_name)\n",
    "\n",
    "# Output hash values\n",
    "print(f\"Original Hash: {original_hash}\")\n",
    "print(f\"Modified Hash: {modified_hash}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original Hash: cf348c92be8c3249369823bd5261607e9cd3197cf67214addf87578b8a433fdf\n",
      "Modified Hash: 739d647be45dcab08339cda63405c0c6acf9cfdc0510d1fb08529810cf1c0672\n"
     ]
    }
   ],
   "source": [
    "from safetensors import safe_open\n",
    "from safetensors.torch import save_file\n",
    "import hashlib\n",
    "\n",
    "filename = \"model-00004-of-00004.safetensors\"\n",
    "modified_model_name = \"modified_model4.safetensors\"\n",
    "\n",
    "# Load the original model\n",
    "with safe_open(filename, framework=\"pt\") as f:\n",
    "    tensors = {key: f.get_tensor(key) for key in f.keys()}\n",
    "\n",
    "# Define anonymous metadata\n",
    "metadata = {\n",
    "    \"description\": \"This model has anonymous metadata\",  # Neutral description\n",
    "    \"created_by\": \"anonymous\",  # Anonymous creator\n",
    "    \"format\": \"pt\"  # Adjust based on actual format needed\n",
    "}\n",
    "\n",
    "# Save the model with new metadata but without altering weights\n",
    "save_file(tensors, modified_model_name, metadata=metadata)\n",
    "\n",
    "# Function to compute file hash\n",
    "def compute_hash(filename):\n",
    "    hasher = hashlib.sha256()\n",
    "    with open(filename, \"rb\") as f:\n",
    "        buf = f.read()\n",
    "        hasher.update(buf)\n",
    "        return hasher.hexdigest()\n",
    "\n",
    "# Compute hashes\n",
    "original_hash = compute_hash(filename)\n",
    "modified_hash = compute_hash(modified_model_name)\n",
    "\n",
    "# Output hash values\n",
    "print(f\"Original Hash: {original_hash}\")\n",
    "print(f\"Modified Hash: {modified_hash}\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.10.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}