{
  "id": "whisper-small",
  "name": "Whisper Small",
  "components": [
    {
      "id": "input-enc",
      "type": "input",
      "name": "mel_features",
      "position": { "x": 140, "y": 50 },
      "params": { "shape": [1, 80, 3000] },
      "inputs": [], "outputs": []
    },
    {
      "id": "audioConv-1",
      "type": "audioConv",
      "name": "conv1",
      "scope": "encoder.stem",
      "position": { "x": 140, "y": 220 },
      "params": { "outChannels": 384, "kernelSize": 3, "stride": 1, "padding": 1 },
      "inputs": [], "outputs": []
    },
    {
      "id": "gelu-1",
      "type": "gelu",
      "name": "gelu_1",
      "scope": "encoder.stem",
      "position": { "x": 140, "y": 390 },
      "params": {},
      "inputs": [], "outputs": []
    },
    {
      "id": "audioConv-2",
      "type": "audioConv",
      "name": "conv2",
      "scope": "encoder.stem",
      "position": { "x": 140, "y": 560 },
      "params": { "outChannels": 384, "kernelSize": 3, "stride": 2, "padding": 1 },
      "inputs": [], "outputs": []
    },
    {
      "id": "gelu-2",
      "type": "gelu",
      "name": "gelu_2",
      "scope": "encoder.stem",
      "position": { "x": 140, "y": 730 },
      "params": {},
      "inputs": [], "outputs": []
    },
    {
      "id": "permute-enc",
      "type": "permute",
      "name": "to_tokens",
      "scope": "encoder.embeddings",
      "position": { "x": 140, "y": 830 },
      "params": { "dims": [0, 2, 1] },
      "inputs": [], "outputs": []
    },
    {
      "id": "positionalEncoding-enc",
      "type": "positionalEncoding",
      "name": "enc_pos_emb",
      "scope": "encoder.embeddings",
      "position": { "x": 140, "y": 900 },
      "params": { "maxLen": 1500, "embedDim": 384 },
      "inputs": [], "outputs": []
    },
    {
      "id": "transformerBlock-enc-1",
      "type": "transformerBlock",
      "name": "enc_block_1",
      "scope": "encoder.layer.0",
      "position": { "x": 140, "y": 1070 },
      "params": { "embedDim": 384, "numHeads": 6, "ffDim": 1536 },
      "inputs": [], "outputs": []
    },
    {
      "id": "transformerBlock-enc-2",
      "type": "transformerBlock",
      "name": "enc_block_2",
      "scope": "encoder.layer.1",
      "position": { "x": 140, "y": 1240 },
      "params": { "embedDim": 384, "numHeads": 6, "ffDim": 1536 },
      "inputs": [], "outputs": []
    },
    {
      "id": "layerNorm-enc",
      "type": "layerNorm",
      "name": "enc_norm",
      "scope": "encoder",
      "position": { "x": 140, "y": 1410 },
      "params": { "normalizedShape": 384 },
      "inputs": [], "outputs": []
    },
    {
      "id": "input-dec",
      "type": "input",
      "name": "decoder_tokens",
      "position": { "x": 460, "y": 50 },
      "params": { "shape": [1, 448] },
      "inputs": [], "outputs": []
    },
    {
      "id": "embedding-dec",
      "type": "embedding",
      "name": "token_embed",
      "scope": "decoder.embeddings",
      "position": { "x": 460, "y": 220 },
      "params": { "numEmbeddings": 51865, "embeddingDim": 384 },
      "inputs": [], "outputs": []
    },
    {
      "id": "positionalEncoding-dec",
      "type": "positionalEncoding",
      "name": "dec_pos_emb",
      "scope": "decoder.embeddings",
      "position": { "x": 460, "y": 390 },
      "params": { "maxLen": 448, "embedDim": 384 },
      "inputs": [], "outputs": []
    },
    {
      "id": "transformerBlock-dec-1",
      "type": "transformerBlock",
      "name": "dec_block_1",
      "scope": "decoder.layer.0",
      "position": { "x": 460, "y": 730 },
      "params": { "embedDim": 384, "numHeads": 6, "ffDim": 1536 },
      "inputs": [], "outputs": []
    },
    {
      "id": "transformerBlock-dec-2",
      "type": "transformerBlock",
      "name": "dec_block_2",
      "scope": "decoder.layer.1",
      "position": { "x": 460, "y": 900 },
      "params": { "embedDim": 384, "numHeads": 6, "ffDim": 1536 },
      "inputs": [], "outputs": []
    },
    {
      "id": "layerNorm-dec",
      "type": "layerNorm",
      "name": "dec_norm",
      "scope": "decoder",
      "position": { "x": 460, "y": 1070 },
      "params": { "normalizedShape": 384 },
      "inputs": [], "outputs": []
    },
    {
      "id": "linear-out",
      "type": "linear",
      "name": "lm_head",
      "scope": "head",
      "position": { "x": 460, "y": 1240 },
      "params": { "outFeatures": 51865 },
      "inputs": [], "outputs": []
    },
    {
      "id": "output-1",
      "type": "output",
      "name": "token_logits",
      "position": { "x": 460, "y": 1410 },
      "params": {},
      "inputs": [], "outputs": []
    }
  ],
  "connections": [
    { "id": "c1",  "from": "input-enc",             "to": "audioConv-1" },
    { "id": "c2",  "from": "audioConv-1",            "to": "gelu-1" },
    { "id": "c3",  "from": "gelu-1",                 "to": "audioConv-2" },
    { "id": "c4",  "from": "audioConv-2",            "to": "gelu-2" },
    { "id": "c5",  "from": "gelu-2",                 "to": "permute-enc" },
    { "id": "c5b", "from": "permute-enc",           "to": "positionalEncoding-enc" },
    { "id": "c6",  "from": "positionalEncoding-enc", "to": "transformerBlock-enc-1" },
    { "id": "c7",  "from": "transformerBlock-enc-1", "to": "transformerBlock-enc-2" },
    { "id": "c8",  "from": "transformerBlock-enc-2", "to": "layerNorm-enc" },
    { "id": "c9",  "from": "input-dec",              "to": "embedding-dec" },
    { "id": "c10", "from": "embedding-dec",          "to": "positionalEncoding-dec" },
    { "id": "c11", "from": "positionalEncoding-dec", "to": "transformerBlock-dec-1" },
    { "id": "c12", "from": "layerNorm-enc",          "to": "transformerBlock-dec-1" },
    { "id": "c13", "from": "transformerBlock-dec-1", "to": "transformerBlock-dec-2" },
    { "id": "c14", "from": "layerNorm-enc",          "to": "transformerBlock-dec-2" },
    { "id": "c15", "from": "transformerBlock-dec-2", "to": "layerNorm-dec" },
    { "id": "c16", "from": "layerNorm-dec",          "to": "linear-out" },
    { "id": "c17", "from": "linear-out",             "to": "output-1" }
  ]
}