Update constants.ts

constants.ts

@@ -1,3 +1,4 @@
+
 import { LayerDefinition, LayerType, GraphTemplate } from './types';
 
 export const LAYER_DEFINITIONS: Record<LayerType, LayerDefinition> = {
@@ -8,7 +9,7 @@ export const LAYER_DEFINITIONS: Record<LayerType, LayerDefinition> = {
     description: 'Entry point for data tensors',
     category: 'Core',
     parameters: [
-      { name: 'modality', type: 'select', label: 'Modality', default: 'Tensor', options: ['Tensor', 'Image', 'Text', 'Audio', 'Video', 'Latent', 'State', '3D Volume'] },
+      { name: 'modality', type: 'select', label: 'Modality', default: 'Tensor', options: ['Tensor', 'Image', 'Text', 'Audio', 'Video', 'Latent', 'State', '3D Volume', 'Point Cloud'] },
       { name: 'shape', type: 'string', label: 'Shape (e.g. 3,224,224)', default: '3, 224, 224' },
       { name: 'batch_size', type: 'number', label: 'Batch Size', default: 32 },
       { name: 'dtype', type: 'select', label: 'Data Type', default: 'float32', options: ['float32', 'int64', 'bool'] }
@@ -58,6 +59,139 @@ export const LAYER_DEFINITIONS: Record<LayerType, LayerDefinition> = {
     ]
   },
 
+  // --- DETECTION (YOLO) ---
+  [LayerType.C2F_BLOCK]: {
+    type: LayerType.C2F_BLOCK,
+    label: 'C2f Block (YOLO)',
+    description: 'CSP Bottleneck with 2 convolutions',
+    category: 'Detection',
+    parameters: [
+      { name: 'c1', type: 'number', label: 'In Channels', default: 64 },
+      { name: 'c2', type: 'number', label: 'Out Channels', default: 64 },
+      { name: 'n', type: 'number', label: 'Num Bottlenecks', default: 1 },
+      { name: 'shortcut', type: 'boolean', label: 'Shortcut', default: true }
+    ]
+  },
+  [LayerType.SPPF_BLOCK]: {
+    type: LayerType.SPPF_BLOCK,
+    label: 'SPPF Block',
+    description: 'Spatial Pyramid Pooling - Fast',
+    category: 'Detection',
+    parameters: [
+      { name: 'c1', type: 'number', label: 'In Channels', default: 512 },
+      { name: 'c2', type: 'number', label: 'Out Channels', default: 512 },
+      { name: 'k', type: 'number', label: 'Kernel Size', default: 5 }
+    ]
+  },
+  [LayerType.DETECT_HEAD]: {
+    type: LayerType.DETECT_HEAD,
+    label: 'Detection Head',
+    description: 'Predicts BBoxes and Classes',
+    category: 'Detection',
+    parameters: [
+      { name: 'nc', type: 'number', label: 'Num Classes', default: 80 },
+      { name: 'ch', type: 'text', label: 'Input Channels', default: '()', description: "List of channels from backbone levels" }
+    ]
+  },
+  [LayerType.ANCHOR_BOX]: {
+    type: LayerType.ANCHOR_BOX,
+    label: 'Anchor Boxes',
+    description: 'Reference boxes for detection',
+    category: 'Detection',
+    parameters: [
+      { name: 'aspect_ratios', type: 'text', label: 'Ratios', default: '[0.5, 1.0, 2.0]' },
+      { name: 'scales', type: 'text', label: 'Scales', default: '[32, 64, 128]' }
+    ]
+  },
+
+  // --- AUDIO / SPEECH ---
+  [LayerType.MEL_SPECTROGRAM]: {
+    type: LayerType.MEL_SPECTROGRAM,
+    label: 'MelSpectrogram',
+    description: 'Raw Audio to Mel Spectrogram',
+    category: 'Audio',
+    parameters: [
+      { name: 'sample_rate', type: 'number', label: 'Sample Rate', default: 22050 },
+      { name: 'n_fft', type: 'number', label: 'N_FFT', default: 1024 },
+      { name: 'n_mels', type: 'number', label: 'Num Mels', default: 80 }
+    ]
+  },
+  [LayerType.CONFORMER_BLOCK]: {
+    type: LayerType.CONFORMER_BLOCK,
+    label: 'Conformer Block',
+    description: 'Convolution + Transformer (ASR)',
+    category: 'Audio',
+    parameters: [
+      { name: 'd_model', type: 'number', label: 'D Model', default: 256 },
+      { name: 'nhead', type: 'number', label: 'Heads', default: 4 },
+      { name: 'kernel_size', type: 'number', label: 'Conv Kernel', default: 31 }
+    ]
+  },
+  [LayerType.WAVENET_BLOCK]: {
+    type: LayerType.WAVENET_BLOCK,
+    label: 'WaveNet Block',
+    description: 'Dilated Causal Convolution',
+    category: 'Audio',
+    parameters: [
+      { name: 'channels', type: 'number', label: 'Channels', default: 64 },
+      { name: 'dilation', type: 'number', label: 'Dilation', default: 1 },
+      { name: 'kernel_size', type: 'number', label: 'Kernel', default: 3 }
+    ]
+  },
+  [LayerType.VOCODER]: {
+    type: LayerType.VOCODER,
+    label: 'Vocoder',
+    description: 'Mel Spectrogram to Waveform',
+    category: 'Audio',
+    parameters: [
+      { name: 'type', type: 'select', label: 'Type', default: 'HiFiGAN', options: ['HiFiGAN', 'WaveGlow', 'MelGAN'] },
+      { name: 'upsample_rates', type: 'text', label: 'Upsample Rates', default: '[8,8,2,2]' }
+    ]
+  },
+  [LayerType.AUDIO_EMBEDDING]: {
+    type: LayerType.AUDIO_EMBEDDING,
+    label: 'Audio Embedding',
+    description: 'Embeddings for discrete audio tokens',
+    category: 'Audio',
+    parameters: [
+      { name: 'num_embeddings', type: 'number', label: 'Num Embeddings', default: 1024 },
+      { name: 'embedding_dim', type: 'number', label: 'Embedding Dim', default: 512 }
+    ]
+  },
+
+  // --- 3D / VISION ---
+  [LayerType.NERF_BLOCK]: {
+    type: LayerType.NERF_BLOCK,
+    label: 'NeRF Block',
+    description: 'MLP for Radiance Fields',
+    category: '3D',
+    parameters: [
+      { name: 'hidden_dim', type: 'number', label: 'Hidden Dim', default: 256 },
+      { name: 'num_layers', type: 'number', label: 'Num Layers', default: 8 },
+      { name: 'skips', type: 'text', label: 'Skip Layers', default: '[4]' }
+    ]
+  },
+  [LayerType.POINTNET_BLOCK]: {
+    type: LayerType.POINTNET_BLOCK,
+    label: 'PointNet Layer',
+    description: 'Point cloud feature extraction',
+    category: '3D',
+    parameters: [
+      { name: 'in_channels', type: 'number', label: 'In Channels', default: 3 },
+      { name: 'out_channels', type: 'number', label: 'Out Channels', default: 64 }
+    ]
+  },
+  [LayerType.GAUSSIAN_SPLAT]: {
+    type: LayerType.GAUSSIAN_SPLAT,
+    label: 'Gaussian Splat',
+    description: '3D Gaussian Splatting Decoder',
+    category: '3D',
+    parameters: [
+      { name: 'num_gaussians', type: 'number', label: 'Num Gaussians', default: 10000 },
+      { name: 'sh_degree', type: 'number', label: 'SH Degree', default: 3 }
+    ]
+  },
+
   // --- MERGE ---
   [LayerType.CONCAT]: {
     type: LayerType.CONCAT,
@@ -493,6 +627,17 @@ export const LAYER_DEFINITIONS: Record<LayerType, LayerDefinition> = {
       { name: 'dropout', type: 'number', label: 'Dropout', default: 0.1 }
     ]
   },
+  [LayerType.WINDOW_ATTENTION]: {
+    type: LayerType.WINDOW_ATTENTION,
+    label: 'Window Attention',
+    description: 'Sliding/Windowed Attention',
+    category: 'Transformer',
+    parameters: [
+      { name: 'window_size', type: 'number', label: 'Window Size', default: 7 },
+      { name: 'embed_dim', type: 'number', label: 'Embed Dim', default: 96 },
+      { name: 'num_heads', type: 'number', label: 'Num Heads', default: 4 }
+    ]
+  },
   [LayerType.TRANSFORMER_ENCODER]: {
     type: LayerType.TRANSFORMER_ENCODER,
     label: 'Encoder Layer',
@@ -558,6 +703,87 @@ export const INITIAL_EDGES = [
 ];
 
 export const TEMPLATES: Record<string, GraphTemplate> = {
+  'yolo_v8': {
+    id: 'yolo_v8',
+    name: 'YOLO (Detection)',
+    description: 'C2f Backbone with Detection Head.',
+    nodes: [
+      { id: 'in', type: 'custom', position: {x: 300, y: 0}, data: {label: 'Image', type: LayerType.INPUT, params: {modality: 'Image'}} },
+      { id: 'c1', type: 'custom', position: {x: 300, y: 100}, data: {label: 'Conv (Stem)', type: LayerType.CONV2D, params: {out_channels: 64, kernel_size: 3, stride: 2}} },
+      { id: 'c2f1', type: 'custom', position: {x: 300, y: 200}, data: {label: 'C2f Block 1', type: LayerType.C2F_BLOCK, params: {c1: 64, c2: 128, n: 3}} },
+      { id: 'c2f2', type: 'custom', position: {x: 300, y: 300}, data: {label: 'C2f Block 2', type: LayerType.C2F_BLOCK, params: {c1: 128, c2: 256, n: 6}} },
+      { id: 'sppf', type: 'custom', position: {x: 300, y: 400}, data: {label: 'SPPF', type: LayerType.SPPF_BLOCK, params: {c1: 256, c2: 256}} },
+      { id: 'head', type: 'custom', position: {x: 300, y: 500}, data: {label: 'YOLO Head', type: LayerType.DETECT_HEAD, params: {nc: 80}} },
+    ],
+    edges: [
+      { id: '1', source: 'in', target: 'c1' }, { id: '2', source: 'c1', target: 'c2f1' },
+      { id: '3', source: 'c2f1', target: 'c2f2' }, { id: '4', source: 'c2f2', target: 'sppf' },
+      { id: '5', source: 'sppf', target: 'head' }
+    ]
+  },
+  'whisper_stt': {
+    id: 'whisper_stt',
+    name: 'Whisper (STT)',
+    description: 'Speech-to-Text Transformer.',
+    nodes: [
+      { id: 'aud', type: 'custom', position: {x: 100, y: 0}, data: {label: 'Audio', type: LayerType.INPUT, params: {modality: 'Audio'}} },
+      { id: 'mel', type: 'custom', position: {x: 100, y: 100}, data: {label: 'MelSpectrogram', type: LayerType.MEL_SPECTROGRAM, params: {n_mels: 80}} },
+      { id: 'conv1', type: 'custom', position: {x: 100, y: 200}, data: {label: 'Conv1D', type: LayerType.CONV1D, params: {out_channels: 512, kernel_size: 3}} },
+      { id: 'enc', type: 'custom', position: {x: 100, y: 300}, data: {label: 'Transformer Enc', type: LayerType.TRANSFORMER_ENCODER, params: {d_model: 512}} },
+      
+      { id: 'tok', type: 'custom', position: {x: 500, y: 0}, data: {label: 'Text Tokens', type: LayerType.INPUT, params: {modality: 'Text'}} },
+      { id: 'emb', type: 'custom', position: {x: 500, y: 100}, data: {label: 'Embedding', type: LayerType.EMBEDDING, params: {}} },
+      
+      { id: 'dec', type: 'custom', position: {x: 300, y: 400}, data: {label: 'Transformer Dec', type: LayerType.TRANSFORMER_DECODER, params: {d_model: 512}} },
+      { id: 'out', type: 'custom', position: {x: 300, y: 500}, data: {label: 'Next Token', type: LayerType.LINEAR, params: {}} }
+    ],
+    edges: [
+      { id: '1', source: 'aud', target: 'mel' }, { id: '2', source: 'mel', target: 'conv1' },
+      { id: '3', source: 'conv1', target: 'enc' }, { id: '4', source: 'enc', target: 'dec' },
+      { id: '5', source: 'tok', target: 'emb' }, { id: '6', source: 'emb', target: 'dec' },
+      { id: '7', source: 'dec', target: 'out' }
+    ]
+  },
+  'tacotron_tts': {
+    id: 'tacotron_tts',
+    name: 'Voice Cloning (TTS)',
+    description: 'Text to Mel Spectrogram with Vocoder.',
+    nodes: [
+      { id: 'txt', type: 'custom', position: {x: 300, y: 0}, data: {label: 'Text', type: LayerType.INPUT, params: {modality: 'Text'}} },
+      { id: 'emb', type: 'custom', position: {x: 300, y: 100}, data: {label: 'Char Embed', type: LayerType.EMBEDDING, params: {embedding_dim: 512}} },
+      { id: 'pre', type: 'custom', position: {x: 300, y: 200}, data: {label: 'PreNet (Linear)', type: LayerType.LINEAR, params: {out_features: 256}} },
+      { id: 'lstm', type: 'custom', position: {x: 300, y: 300}, data: {label: 'Decoder LSTM', type: LayerType.LSTM, params: {hidden_size: 1024}} },
+      { id: 'mel', type: 'custom', position: {x: 300, y: 400}, data: {label: 'Mel Projection', type: LayerType.LINEAR, params: {out_features: 80}} },
+      { id: 'voc', type: 'custom', position: {x: 300, y: 500}, data: {label: 'HiFiGAN Vocoder', type: LayerType.VOCODER, params: {type: 'HiFiGAN'}} },
+      { id: 'out', type: 'custom', position: {x: 300, y: 600}, data: {label: 'Audio Waveform', type: LayerType.OUTPUT, params: {}} }
+    ],
+    edges: [
+      { id: '1', source: 'txt', target: 'emb' }, { id: '2', source: 'emb', target: 'pre' },
+      { id: '3', source: 'pre', target: 'lstm' }, { id: '4', source: 'lstm', target: 'mel' },
+      { id: '5', source: 'mel', target: 'voc' }, { id: '6', source: 'voc', target: 'out' }
+    ]
+  },
+  'nerf_3d': {
+    id: 'nerf_3d',
+    name: 'NeRF (3D Gen)',
+    description: 'Neural Radiance Field MLP.',
+    nodes: [
+      { id: 'pos', type: 'custom', position: {x: 200, y: 0}, data: {label: 'XYZ Coords', type: LayerType.INPUT, params: {shape: '3'}} },
+      { id: 'dir', type: 'custom', position: {x: 400, y: 0}, data: {label: 'View Dir', type: LayerType.INPUT, params: {shape: '3'}} },
+      { id: 'pe1', type: 'custom', position: {x: 200, y: 100}, data: {label: 'Pos Enc', type: LayerType.POS_EMBED, params: {}} },
+      { id: 'pe2', type: 'custom', position: {x: 400, y: 100}, data: {label: 'Dir Enc', type: LayerType.POS_EMBED, params: {}} },
+      { id: 'mlp1', type: 'custom', position: {x: 300, y: 200}, data: {label: 'Density MLP', type: LayerType.NERF_BLOCK, params: {hidden_dim: 256, num_layers: 8}} },
+      { id: 'cat', type: 'custom', position: {x: 300, y: 300}, data: {label: 'Concat View', type: LayerType.CONCAT, params: {}} },
+      { id: 'mlp2', type: 'custom', position: {x: 300, y: 400}, data: {label: 'Color MLP', type: LayerType.LINEAR, params: {out_features: 128}} },
+      { id: 'rgb', type: 'custom', position: {x: 300, y: 500}, data: {label: 'RGB + Sigma', type: LayerType.OUTPUT, params: {}} }
+    ],
+    edges: [
+      { id: '1', source: 'pos', target: 'pe1' }, { id: '2', source: 'dir', target: 'pe2' },
+      { id: '3', source: 'pe1', target: 'mlp1' }, { id: '4', source: 'mlp1', target: 'cat' },
+      { id: '5', source: 'pe2', target: 'cat' }, { id: '6', source: 'cat', target: 'mlp2' },
+      { id: '7', source: 'mlp2', target: 'rgb' }
+    ]
+  },
   'gpt_style': {
     id: 'gpt_style',
     name: 'LLM (GPT Style)',
@@ -778,4 +1004,4 @@ export const TEMPLATES: Record<string, GraphTemplate> = {
       { id: '4', source: 'cat', target: 'unet' }, { id: '5', source: 'unet', target: 'out' }
     ]
   }
-};
+};