{
  "id": "clip-vit-b32",
  "name": "CLIP ViT-B/32",
  "description": "CLIP ViT-B/32 (OpenAI): full graph; dimensions verified against the official config.json.",
  "components": [
    {
      "id": "n1",
      "type": "input",
      "name": "image",
      "position": {
        "x": 150,
        "y": 50
      },
      "params": {
        "shape": [
          3,
          224,
          224
        ]
      },
      "inputs": [],
      "outputs": [
        "n2"
      ]
    },
    {
      "id": "n2",
      "type": "patchEmbed",
      "name": "patch_embed",
      "position": {
        "x": 150,
        "y": 200
      },
      "params": {
        "imgSize": 224,
        "patchSize": 32,
        "embedDim": 768,
        "inChans": 3
      },
      "inputs": [
        "n1"
      ],
      "outputs": [
        "n3"
      ],
      "scope": "vision",
      "notes": "32x32 patches: 224x224 image becomes 49 tokens of 768 dims (plus a class token)."
    },
    {
      "id": "n3",
      "type": "positionalEncoding",
      "name": "vis_pos",
      "position": {
        "x": 150,
        "y": 350
      },
      "params": {
        "maxLen": 50,
        "embedDim": 768
      },
      "inputs": [
        "n2"
      ],
      "outputs": [
        "n4"
      ],
      "scope": "vision"
    },
    {
      "id": "n4",
      "type": "transformerBlock",
      "name": "vis_block_1",
      "position": {
        "x": 150,
        "y": 500
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n3"
      ],
      "outputs": [
        "n5"
      ],
      "scope": "vision.layer.0",
      "notes": "Pre-norm ViT block (12 heads, MLP 3072, QuickGELU). One of 12."
    },
    {
      "id": "n5",
      "type": "transformerBlock",
      "name": "vis_block_2",
      "position": {
        "x": 150,
        "y": 650
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n4"
      ],
      "outputs": [
        "n6"
      ],
      "scope": "vision.layer.1"
    },
    {
      "id": "n6",
      "type": "transformerBlock",
      "name": "vis_block_3",
      "position": {
        "x": 150,
        "y": 800
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n5"
      ],
      "outputs": [
        "n7"
      ],
      "scope": "vision.layer.2"
    },
    {
      "id": "n7",
      "type": "transformerBlock",
      "name": "vis_block_4",
      "position": {
        "x": 150,
        "y": 950
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n6"
      ],
      "outputs": [
        "n8"
      ],
      "scope": "vision.layer.3"
    },
    {
      "id": "n8",
      "type": "transformerBlock",
      "name": "vis_block_5",
      "position": {
        "x": 150,
        "y": 1100
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n7"
      ],
      "outputs": [
        "n9"
      ],
      "scope": "vision.layer.4"
    },
    {
      "id": "n9",
      "type": "transformerBlock",
      "name": "vis_block_6",
      "position": {
        "x": 150,
        "y": 1250
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n8"
      ],
      "outputs": [
        "n10"
      ],
      "scope": "vision.layer.5"
    },
    {
      "id": "n10",
      "type": "transformerBlock",
      "name": "vis_block_7",
      "position": {
        "x": 150,
        "y": 1400
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n9"
      ],
      "outputs": [
        "n11"
      ],
      "scope": "vision.layer.6"
    },
    {
      "id": "n11",
      "type": "transformerBlock",
      "name": "vis_block_8",
      "position": {
        "x": 150,
        "y": 1550
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n10"
      ],
      "outputs": [
        "n12"
      ],
      "scope": "vision.layer.7"
    },
    {
      "id": "n12",
      "type": "transformerBlock",
      "name": "vis_block_9",
      "position": {
        "x": 150,
        "y": 1700
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n11"
      ],
      "outputs": [
        "n13"
      ],
      "scope": "vision.layer.8"
    },
    {
      "id": "n13",
      "type": "transformerBlock",
      "name": "vis_block_10",
      "position": {
        "x": 150,
        "y": 1850
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n12"
      ],
      "outputs": [
        "n14"
      ],
      "scope": "vision.layer.9"
    },
    {
      "id": "n14",
      "type": "transformerBlock",
      "name": "vis_block_11",
      "position": {
        "x": 150,
        "y": 2000
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n13"
      ],
      "outputs": [
        "n15"
      ],
      "scope": "vision.layer.10"
    },
    {
      "id": "n15",
      "type": "transformerBlock",
      "name": "vis_block_12",
      "position": {
        "x": 150,
        "y": 2150
      },
      "params": {
        "embedDim": 768,
        "numHeads": 12,
        "ffDim": 3072
      },
      "inputs": [
        "n14"
      ],
      "outputs": [
        "n16"
      ],
      "scope": "vision.layer.11"
    },
    {
      "id": "n16",
      "type": "layerNorm",
      "name": "vis_norm",
      "position": {
        "x": 150,
        "y": 2300
      },
      "params": {
        "normalizedShape": 768
      },
      "inputs": [
        "n15"
      ],
      "outputs": [
        "n17"
      ],
      "scope": "vision"
    },
    {
      "id": "n17",
      "type": "globalAvgPool1d",
      "name": "vis_pool",
      "position": {
        "x": 150,
        "y": 2450
      },
      "params": {},
      "inputs": [
        "n16"
      ],
      "outputs": [
        "n18"
      ],
      "scope": "vision",
      "notes": "Stand-in for class-token selection."
    },
    {
      "id": "n18",
      "type": "linear",
      "name": "img_proj",
      "position": {
        "x": 150,
        "y": 2600
      },
      "params": {
        "inFeatures": 768,
        "outFeatures": 512,
        "bias": false
      },
      "inputs": [
        "n17"
      ],
      "outputs": [
        "n37"
      ],
      "scope": "vision",
      "notes": "Projection into the shared 512-dim embedding space."
    },
    {
      "id": "n19",
      "type": "input",
      "name": "tokens",
      "position": {
        "x": 650,
        "y": 50
      },
      "params": {
        "shape": [
          1,
          77
        ]
      },
      "inputs": [],
      "outputs": [
        "n20"
      ],
      "notes": "77-token max context."
    },
    {
      "id": "n20",
      "type": "embedding",
      "name": "tok_embed",
      "position": {
        "x": 650,
        "y": 200
      },
      "params": {
        "numEmbeddings": 49408,
        "embeddingDim": 512
      },
      "inputs": [
        "n19"
      ],
      "outputs": [
        "n21"
      ],
      "scope": "text",
      "notes": "49408-token BPE vocabulary."
    },
    {
      "id": "n21",
      "type": "positionalEncoding",
      "name": "txt_pos",
      "position": {
        "x": 650,
        "y": 350
      },
      "params": {
        "maxLen": 77,
        "embedDim": 512
      },
      "inputs": [
        "n20"
      ],
      "outputs": [
        "n22"
      ],
      "scope": "text"
    },
    {
      "id": "n22",
      "type": "transformerBlock",
      "name": "txt_block_1",
      "position": {
        "x": 650,
        "y": 500
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n21"
      ],
      "outputs": [
        "n23"
      ],
      "scope": "text.layer.0",
      "notes": "Causal text block (8 heads, MLP 2048, QuickGELU). One of 12."
    },
    {
      "id": "n23",
      "type": "transformerBlock",
      "name": "txt_block_2",
      "position": {
        "x": 650,
        "y": 650
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n22"
      ],
      "outputs": [
        "n24"
      ],
      "scope": "text.layer.1"
    },
    {
      "id": "n24",
      "type": "transformerBlock",
      "name": "txt_block_3",
      "position": {
        "x": 650,
        "y": 800
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n23"
      ],
      "outputs": [
        "n25"
      ],
      "scope": "text.layer.2"
    },
    {
      "id": "n25",
      "type": "transformerBlock",
      "name": "txt_block_4",
      "position": {
        "x": 650,
        "y": 950
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n24"
      ],
      "outputs": [
        "n26"
      ],
      "scope": "text.layer.3"
    },
    {
      "id": "n26",
      "type": "transformerBlock",
      "name": "txt_block_5",
      "position": {
        "x": 650,
        "y": 1100
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n25"
      ],
      "outputs": [
        "n27"
      ],
      "scope": "text.layer.4"
    },
    {
      "id": "n27",
      "type": "transformerBlock",
      "name": "txt_block_6",
      "position": {
        "x": 650,
        "y": 1250
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n26"
      ],
      "outputs": [
        "n28"
      ],
      "scope": "text.layer.5"
    },
    {
      "id": "n28",
      "type": "transformerBlock",
      "name": "txt_block_7",
      "position": {
        "x": 650,
        "y": 1400
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n27"
      ],
      "outputs": [
        "n29"
      ],
      "scope": "text.layer.6"
    },
    {
      "id": "n29",
      "type": "transformerBlock",
      "name": "txt_block_8",
      "position": {
        "x": 650,
        "y": 1550
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n28"
      ],
      "outputs": [
        "n30"
      ],
      "scope": "text.layer.7"
    },
    {
      "id": "n30",
      "type": "transformerBlock",
      "name": "txt_block_9",
      "position": {
        "x": 650,
        "y": 1700
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n29"
      ],
      "outputs": [
        "n31"
      ],
      "scope": "text.layer.8"
    },
    {
      "id": "n31",
      "type": "transformerBlock",
      "name": "txt_block_10",
      "position": {
        "x": 650,
        "y": 1850
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n30"
      ],
      "outputs": [
        "n32"
      ],
      "scope": "text.layer.9"
    },
    {
      "id": "n32",
      "type": "transformerBlock",
      "name": "txt_block_11",
      "position": {
        "x": 650,
        "y": 2000
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n31"
      ],
      "outputs": [
        "n33"
      ],
      "scope": "text.layer.10"
    },
    {
      "id": "n33",
      "type": "transformerBlock",
      "name": "txt_block_12",
      "position": {
        "x": 650,
        "y": 2150
      },
      "params": {
        "embedDim": 512,
        "numHeads": 8,
        "ffDim": 2048
      },
      "inputs": [
        "n32"
      ],
      "outputs": [
        "n34"
      ],
      "scope": "text.layer.11"
    },
    {
      "id": "n34",
      "type": "layerNorm",
      "name": "txt_norm",
      "position": {
        "x": 650,
        "y": 2300
      },
      "params": {
        "normalizedShape": 512
      },
      "inputs": [
        "n33"
      ],
      "outputs": [
        "n35"
      ],
      "scope": "text"
    },
    {
      "id": "n35",
      "type": "globalAvgPool1d",
      "name": "txt_pool",
      "position": {
        "x": 650,
        "y": 2450
      },
      "params": {},
      "inputs": [
        "n34"
      ],
      "outputs": [
        "n36"
      ],
      "scope": "text",
      "notes": "Stand-in for EOT-token selection."
    },
    {
      "id": "n36",
      "type": "linear",
      "name": "txt_proj",
      "position": {
        "x": 650,
        "y": 2600
      },
      "params": {
        "inFeatures": 512,
        "outFeatures": 512,
        "bias": false
      },
      "inputs": [
        "n35"
      ],
      "outputs": [
        "n37"
      ],
      "scope": "text",
      "notes": "Projection into the shared 512-dim embedding space."
    },
    {
      "id": "n37",
      "type": "matmul",
      "name": "similarity",
      "position": {
        "x": 400,
        "y": 2750
      },
      "params": {},
      "inputs": [
        "n18",
        "n36"
      ],
      "outputs": [
        "n38"
      ],
      "notes": "Scaled dot product between L2-normalized image and text embeddings (learned temperature). Over a batch this is the NxN contrastive logit matrix."
    },
    {
      "id": "n38",
      "type": "output",
      "name": "logits",
      "position": {
        "x": 400,
        "y": 2900
      },
      "params": {},
      "inputs": [
        "n37"
      ],
      "outputs": []
    }
  ],
  "connections": [
    {
      "id": "c1",
      "from": "n1",
      "to": "n2"
    },
    {
      "id": "c2",
      "from": "n2",
      "to": "n3"
    },
    {
      "id": "c3",
      "from": "n3",
      "to": "n4"
    },
    {
      "id": "c4",
      "from": "n4",
      "to": "n5"
    },
    {
      "id": "c5",
      "from": "n5",
      "to": "n6"
    },
    {
      "id": "c6",
      "from": "n6",
      "to": "n7"
    },
    {
      "id": "c7",
      "from": "n7",
      "to": "n8"
    },
    {
      "id": "c8",
      "from": "n8",
      "to": "n9"
    },
    {
      "id": "c9",
      "from": "n9",
      "to": "n10"
    },
    {
      "id": "c10",
      "from": "n10",
      "to": "n11"
    },
    {
      "id": "c11",
      "from": "n11",
      "to": "n12"
    },
    {
      "id": "c12",
      "from": "n12",
      "to": "n13"
    },
    {
      "id": "c13",
      "from": "n13",
      "to": "n14"
    },
    {
      "id": "c14",
      "from": "n14",
      "to": "n15"
    },
    {
      "id": "c15",
      "from": "n15",
      "to": "n16"
    },
    {
      "id": "c16",
      "from": "n16",
      "to": "n17"
    },
    {
      "id": "c17",
      "from": "n17",
      "to": "n18"
    },
    {
      "id": "c18",
      "from": "n19",
      "to": "n20"
    },
    {
      "id": "c19",
      "from": "n20",
      "to": "n21"
    },
    {
      "id": "c20",
      "from": "n21",
      "to": "n22"
    },
    {
      "id": "c21",
      "from": "n22",
      "to": "n23"
    },
    {
      "id": "c22",
      "from": "n23",
      "to": "n24"
    },
    {
      "id": "c23",
      "from": "n24",
      "to": "n25"
    },
    {
      "id": "c24",
      "from": "n25",
      "to": "n26"
    },
    {
      "id": "c25",
      "from": "n26",
      "to": "n27"
    },
    {
      "id": "c26",
      "from": "n27",
      "to": "n28"
    },
    {
      "id": "c27",
      "from": "n28",
      "to": "n29"
    },
    {
      "id": "c28",
      "from": "n29",
      "to": "n30"
    },
    {
      "id": "c29",
      "from": "n30",
      "to": "n31"
    },
    {
      "id": "c30",
      "from": "n31",
      "to": "n32"
    },
    {
      "id": "c31",
      "from": "n32",
      "to": "n33"
    },
    {
      "id": "c32",
      "from": "n33",
      "to": "n34"
    },
    {
      "id": "c33",
      "from": "n34",
      "to": "n35"
    },
    {
      "id": "c34",
      "from": "n35",
      "to": "n36"
    },
    {
      "id": "c35",
      "from": "n18",
      "to": "n37"
    },
    {
      "id": "c36",
      "from": "n36",
      "to": "n37"
    },
    {
      "id": "c37",
      "from": "n37",
      "to": "n38"
    }
  ]
}