Upload sd_token_similarity_calculator.ipynb

sd_token_similarity_calculator.ipynb

@@ -17,7 +17,7 @@
     {
       "cell_type": "markdown",
       "source": [
-        "This Notebook is a Stable-diffusion tool which allows you to find similiar tokens from the SD 1.5 vocab.json that you can use for text-to-image generation"
+        "This Notebook is a Stable-diffusion tool which allows you to find similiar tokens from the SD 1.5 vocab.json that you can use for text-to-image generation."
       ],
       "metadata": {
         "id": "L7JTcbOdBPfh"
@@ -101,13 +101,15 @@
     {
       "cell_type": "code",
       "source": [
-        "\n",
         "from transformers import AutoTokenizer\n",
         "tokenizer = AutoTokenizer.from_pretrained(\"openai/clip-vit-large-patch14\", clean_up_tokenization_spaces = False)\n",
         "prompt= \"banana\" # @param {type:'string'}\n",
         "tokenizer_output = tokenizer(text = prompt)\n",
         "input_ids = tokenizer_output['input_ids']\n",
-        "print(input_ids)"
+        "print(input_ids)\n",
+        "id_A = input_ids[1]\n",
+        "A = token[id_A]\n",
+        "_A = LA.vector_norm(A, ord=2)"
       ],
       "metadata": {
         "id": "RPdkYzT2_X85"
@@ -115,16 +117,62 @@
       "execution_count": null,
       "outputs": []
     },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "OPTIONAL : Add/subtract + normalize above result with another token"
+      ],
+      "metadata": {
+        "id": "JKnz0aLFVGXc"
+      }
+    },
     {
       "cell_type": "code",
       "source": [
-        "#Produce a list id IDs that are most similiar to the prompt ID at positiion 1\n",
+        "mix_with = \"\" # @param {type:'string'}\n",
+        "mix_method = 'None' # @param [\"None\" , \"Average\", \"Subtract\"] {allow-input: true}\n",
+        "w = 0.5 # @param {type:\"slider\", min:0, max:1, step:0.01}\n",
         "\n",
-        "id_A = input_ids[1]\n",
-        "A = token[id_A]\n",
-        "_A = LA.vector_norm(A, ord=2)\n",
-        "dots = torch.zeros(NUM_TOKENS)\n",
         "\n",
+        "\n",
+        "tokenizer_output = tokenizer(text = mix_with)\n",
+        "input_ids = tokenizer_output['input_ids']\n",
+        "id_C = input_ids[1]\n",
+        "C = token[id_C]\n",
+        "_C = LA.vector_norm(C, ord=2)\n",
+        "\n",
+        "if (mix_method ==  \"Average\"):\n",
+        "  A = w*A + (1-w)*C\n",
+        "  _A = LA.vector_norm(A, ord=2)\n",
+        "\n",
+        "if (mix_method ==  \"Subtract\"):\n",
+        "  tmp = w*A - (1-w)*C\n",
+        "  _tmp = LA.vector_norm(tmp, ord=2)\n",
+        "  A = tmp*((w*_A + (1-w)*_C)/_tmp)\n",
+        "  _A = LA.vector_norm(A, ord=2)\n",
+        "\n",
+        "\n"
+      ],
+      "metadata": {
+        "id": "oXbNSRSKPgRr"
+      },
+      "execution_count": 6,
+      "outputs": []
+    },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "Produce a list id IDs that are most similiar to the prompt ID at positiion 1 based on above result"
+      ],
+      "metadata": {
+        "id": "3uBSZ1vWVCew"
+      }
+    },
+    {
+      "cell_type": "code",
+      "source": [
+        "\n",
+        "dots = torch.zeros(NUM_TOKENS)\n",
         "for index in range(NUM_TOKENS):\n",
         "  id_B = index\n",
         "  B = token[id_B]\n",
@@ -135,8 +183,12 @@
         "\n",
         "sorted, indices = torch.sort(dots,dim=0 , descending=True)\n",
         "#----#\n",
-        "print(f'Calculated all cosine-similarities between the token {vocab[id_A]} with ID = {id_A} the rest of the {NUM_TOKENS} tokens as a 1x{sorted.shape[0]} tensor')\n",
-        "print(f'Calculated indices as a 1x{indices.shape[0]} tensor')"
+        "if (mix_method ==  \"Average\"):\n",
+        "  print(f'Calculated all cosine-similarities between the average of token {vocab[id_A]} and {vocab[id_C]} with ID = {id_A} and mixed ID = {id_C} as a 1x{sorted.shape[0]} tensor')\n",
+        "if (mix_method ==  \"Subtract\"):\n",
+        "  print(f'Calculated all cosine-similarities between the subtract of token {vocab[id_A]} and {vocab[id_C]} with ID = {id_A} and mixed ID = {id_C} as a 1x{sorted.shape[0]} tensor')\n",
+        "if (mix_method ==  \"None\"):\n",
+        "  print(f'Calculated all cosine-similarities between the token {vocab[id_A]} with ID = {id_A} the rest of the {NUM_TOKENS} tokens as a 1x{sorted.shape[0]} tensor')"
       ],
       "metadata": {
         "id": "juxsvco9B0iV"
@@ -144,6 +196,15 @@
       "execution_count": null,
       "outputs": []
     },
+    {
+      "cell_type": "markdown",
+      "source": [
+        "Print the sorted list from above result"
+      ],
+      "metadata": {
+        "id": "y-Ig3glrVQC3"
+      }
+    },
     {
       "cell_type": "code",
       "source": [
@@ -152,7 +213,7 @@
         "print_ID = False # @param {type:\"boolean\"}\n",
         "print_Similarity = True # @param {type:\"boolean\"}\n",
         "print_Name = True # @param {type:\"boolean\"}\n",
-        "print_Divider = False # @param {type:\"boolean\"}\n",
+        "print_Divider = True # @param {type:\"boolean\"}\n",
         "\n",
         "for index in range(list_size):\n",
         "  id = indices[index].item()\n",
@@ -166,10 +227,221 @@
         "    print('--------')"
       ],
       "metadata": {
-        "id": "YIEmLAzbHeuo"
+        "id": "YIEmLAzbHeuo",
+        "outputId": "843fbd7c-b208-49e0-9793-69bb36622c27",
+        "colab": {
+          "base_uri": "https://localhost:8080/"
+        }
       },
-      "execution_count": null,
-      "outputs": []
+      "execution_count": 5,
+      "outputs": [
+        {
+          "output_type": "stream",
+          "name": "stdout",
+          "text": [
+            "banana</w>\n",
+            "similiarity = 74.26 %\n",
+            "nude</w>\n",
+            "similiarity = 72.49 %\n",
+            "bananas</w>\n",
+            "similiarity = 30.34 %\n",
+            "nudes</w>\n",
+            "similiarity = 27.19 %\n",
+            "banan\n",
+            "similiarity = 25.08 %\n",
+            "ðŁįĮ</w>\n",
+            "similiarity = 22.27 %\n",
+            "naked</w>\n",
+            "similiarity = 22.12 %\n",
+            "orange</w>\n",
+            "similiarity = 19.53 %\n",
+            "cucumber</w>\n",
+            "similiarity = 17.36 %\n",
+            "nutella</w>\n",
+            "similiarity = 17.33 %\n",
+            "camel</w>\n",
+            "similiarity = 17.22 %\n",
+            "eggplant</w>\n",
+            "similiarity = 17.13 %\n",
+            "swimsuit</w>\n",
+            "similiarity = 16.62 %\n",
+            "chicken</w>\n",
+            "similiarity = 16.38 %\n",
+            "bikini</w>\n",
+            "similiarity = 16.08 %\n",
+            "grape</w>\n",
+            "similiarity = 16.01 %\n",
+            "ballerina</w>\n",
+            "similiarity = 16.01 %\n",
+            "mango</w>\n",
+            "similiarity = 16.0 %\n",
+            "manicure</w>\n",
+            "similiarity = 15.8 %\n",
+            "pencil</w>\n",
+            "similiarity = 15.62 %\n",
+            "yoga</w>\n",
+            "similiarity = 15.56 %\n",
+            "indian</w>\n",
+            "similiarity = 15.51 %\n",
+            "yellow</w>\n",
+            "similiarity = 15.51 %\n",
+            "venus</w>\n",
+            "similiarity = 15.5 %\n",
+            "snake</w>\n",
+            "similiarity = 15.41 %\n",
+            "dunk</w>\n",
+            "similiarity = 15.39 %\n",
+            "ters\n",
+            "similiarity = 15.27 %\n",
+            "underwear</w>\n",
+            "similiarity = 15.26 %\n",
+            "sunbathing</w>\n",
+            "similiarity = 15.15 %\n",
+            "potato</w>\n",
+            "similiarity = 15.04 %\n",
+            "milk</w>\n",
+            "similiarity = 14.91 %\n",
+            "bamboo</w>\n",
+            "similiarity = 14.85 %\n",
+            "selfie</w>\n",
+            "similiarity = 14.85 %\n",
+            "features</w>\n",
+            "similiarity = 14.82 %\n",
+            "know\n",
+            "similiarity = 14.79 %\n",
+            "oilpainting</w>\n",
+            "similiarity = 14.7 %\n",
+            "reas\n",
+            "similiarity = 14.63 %\n",
+            "croissant</w>\n",
+            "similiarity = 14.61 %\n",
+            "oranges</w>\n",
+            "similiarity = 14.59 %\n",
+            "conversation</w>\n",
+            "similiarity = 14.57 %\n",
+            "photoshoot</w>\n",
+            "similiarity = 14.55 %\n",
+            "ery\n",
+            "similiarity = 14.49 %\n",
+            "pear</w>\n",
+            "similiarity = 14.42 %\n",
+            "mcnam\n",
+            "similiarity = 14.42 %\n",
+            "dens</w>\n",
+            "similiarity = 14.38 %\n",
+            "cigarette</w>\n",
+            "similiarity = 14.33 %\n",
+            "tangerine</w>\n",
+            "similiarity = 14.3 %\n",
+            "aluminum</w>\n",
+            "similiarity = 14.28 %\n",
+            "plum</w>\n",
+            "similiarity = 14.28 %\n",
+            "rape</w>\n",
+            "similiarity = 14.24 %\n",
+            "apple</w>\n",
+            "similiarity = 14.2 %\n",
+            "apd</w>\n",
+            "similiarity = 14.17 %\n",
+            "safari</w>\n",
+            "similiarity = 14.09 %\n",
+            "yolo</w>\n",
+            "similiarity = 14.06 %\n",
+            "hoodie</w>\n",
+            "similiarity = 13.96 %\n",
+            "cabaret</w>\n",
+            "similiarity = 13.91 %\n",
+            "superman</w>\n",
+            "similiarity = 13.9 %\n",
+            "saree</w>\n",
+            "similiarity = 13.86 %\n",
+            "mommy</w>\n",
+            "similiarity = 13.78 %\n",
+            "sausage</w>\n",
+            "similiarity = 13.76 %\n",
+            "marshmallow</w>\n",
+            "similiarity = 13.75 %\n",
+            "latex</w>\n",
+            "similiarity = 13.74 %\n",
+            "blonde</w>\n",
+            "similiarity = 13.69 %\n",
+            "champagne</w>\n",
+            "similiarity = 13.62 %\n",
+            "parachute</w>\n",
+            "similiarity = 13.61 %\n",
+            "stor</w>\n",
+            "similiarity = 13.58 %\n",
+            "feminine</w>\n",
+            "similiarity = 13.55 %\n",
+            "ayu</w>\n",
+            "similiarity = 13.5 %\n",
+            "âĢ¼ï¸ı</w>\n",
+            "similiarity = 13.45 %\n",
+            "naked\n",
+            "similiarity = 13.45 %\n",
+            "poop</w>\n",
+            "similiarity = 13.44 %\n",
+            "honeymoon</w>\n",
+            "similiarity = 13.41 %\n",
+            "giraffe</w>\n",
+            "similiarity = 13.37 %\n",
+            "zebra</w>\n",
+            "similiarity = 13.35 %\n",
+            "mud</w>\n",
+            "similiarity = 13.35 %\n",
+            "blanket</w>\n",
+            "similiarity = 13.34 %\n",
+            "silly</w>\n",
+            "similiarity = 13.32 %\n",
+            "animal</w>\n",
+            "similiarity = 13.31 %\n",
+            "malayalam</w>\n",
+            "similiarity = 13.25 %\n",
+            "mustache</w>\n",
+            "similiarity = 13.25 %\n",
+            "mrc</w>\n",
+            "similiarity = 13.24 %\n",
+            "yuri</w>\n",
+            "similiarity = 13.23 %\n",
+            "japanese</w>\n",
+            "similiarity = 13.19 %\n",
+            "gibbs</w>\n",
+            "similiarity = 13.16 %\n",
+            "ðŁĻĤ\n",
+            "similiarity = 13.15 %\n",
+            "rhubarb</w>\n",
+            "similiarity = 13.14 %\n",
+            "trac\n",
+            "similiarity = 13.13 %\n",
+            "polaroid</w>\n",
+            "similiarity = 13.08 %\n",
+            "lunch</w>\n",
+            "similiarity = 13.04 %\n",
+            "sandal</w>\n",
+            "similiarity = 13.03 %\n",
+            "popart</w>\n",
+            "similiarity = 13.02 %\n",
+            "kissing</w>\n",
+            "similiarity = 13.02 %\n",
+            "funeral</w>\n",
+            "similiarity = 13.02 %\n",
+            "runway</w>\n",
+            "similiarity = 13.01 %\n",
+            "milk\n",
+            "similiarity = 12.98 %\n",
+            "tutu</w>\n",
+            "similiarity = 12.96 %\n",
+            "flag</w>\n",
+            "similiarity = 12.95 %\n",
+            "hours</w>\n",
+            "similiarity = 12.95 %\n",
+            "monet</w>\n",
+            "similiarity = 12.91 %\n",
+            "ali</w>\n",
+            "similiarity = 12.89 %\n"
+          ]
+        }
+      ]
     },
     {
       "cell_type": "markdown",