{ "format": "tensorviz.graph", "version": 3, "exportedAt": "2026-09-17T05:43:00.236026+00:00", "status": "static", "warningCount": 0, "document": { "schemaVersion": 2, "semantic": { "graph_id": "snapshot", "name": "TinyDecoder", "revision": 0, "nodes": [ { "node_id": "input", "kind": "input", "op": "Input", "params": { "shape": "1,6" }, "in_ports": [], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "input", "name": "Input", "displayName": "Input", "description": "Model input node", "icon": "ArrowRight", "category": "io", "parameters": [ { "name": "shape", "type": "string", "defaultValue": "1,3,224,224", "required": true, "description": "Input tensor shape (comma-separated, e.g., 1,3,224,224)" }, { "name": "name", "type": "string", "defaultValue": "input", "required": false, "description": "Name for this input" } ] } } }, { "node_id": "embedding", "kind": "layer", "op": "Embedding", "params": { "num_embeddings": 32, "embedding_dim": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "embedding", "name": "Embedding", "displayName": "Embedding", "description": "Lookup table for dense vector embeddings", "icon": "Hash", "category": "embedding", "parameters": [ { "name": "num_embeddings", "type": "number", "defaultValue": 10000, "required": true, "description": "Size of the dictionary of embeddings", "min": 1 }, { "name": "embedding_dim", "type": "number", "defaultValue": 128, "required": true, "description": "Size of each embedding vector", "min": 1 }, { "name": "padding_idx", "type": "number", "defaultValue": 0, "required": false, "description": "Pad output with zeros at this index", "min": 0 }, { "name": "max_norm", "type": "number", "defaultValue": 0, "required": false, "description": "If > 0, re-normalizes embeddings with norm > max_norm", "min": 0 }, { "name": "sparse", "type": "boolean", "defaultValue": false, "required": false, "description": "If True, gradient w.r.t. weight is sparse" } ] } } }, { "node_id": "positions", "kind": "module", "op": "Module", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "module", "name": "Module", "displayName": "Module", "description": "Custom module group (nn.Module)", "icon": "Package", "category": "module", "parameters": [ { "name": "name", "type": "string", "defaultValue": "CustomModule", "required": true, "description": "Module class name" }, { "name": "description", "type": "string", "defaultValue": "", "required": false, "description": "Module description" } ] }, "moduleInstanceId": "positions", "moduleTemplateId": "model::SinusoidalPositions", "isGroup": true, "moduleMembers": [ "python@positions#1", "python@positions#0", "mul@positions#0", "stack@positions#0", "flatten@positions#0", "add@positions#0" ], "moduleLayerCount": 6 } }, { "node_id": "block", "kind": "module", "op": "Module", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "module", "name": "Module", "displayName": "Module", "description": "Custom module group (nn.Module)", "icon": "Package", "category": "module", "parameters": [ { "name": "name", "type": "string", "defaultValue": "CustomModule", "required": true, "description": "Module class name" }, { "name": "description", "type": "string", "defaultValue": "", "required": false, "description": "Module description" } ] }, "moduleInstanceId": "block", "moduleTemplateId": "model::DecoderBlock", "isGroup": true, "moduleMembers": [ "block.attention", "add@block#0", "block.norm_attention", "block.ffn", "add@block#1", "block.norm_ffn" ], "moduleLayerCount": 18 } }, { "node_id": "vocabulary", "kind": "layer", "op": "Linear", "params": { "in_features": 16, "out_features": 32 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] } } }, { "node_id": "probabilities", "kind": "layer", "op": "Softmax", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "softmax", "name": "Softmax", "displayName": "Softmax", "description": "Softmax over a dimension", "icon": "TrendingUp", "category": "activation", "parameters": [ { "name": "dim", "type": "number", "defaultValue": 1, "required": true, "description": "Dimension along which Softmax is computed" } ] } } }, { "node_id": "output", "kind": "output", "op": "Output", "params": {}, "in_ports": [ "x" ], "out_ports": [], "metadata": { "layerType": { "id": "output", "name": "Output", "displayName": "Output", "description": "Model output node", "icon": "ArrowLeft", "category": "io", "parameters": [ { "name": "name", "type": "string", "defaultValue": "output", "required": false, "description": "Name for this output" } ] } } }, { "node_id": "python@positions#1", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "positions" } }, { "node_id": "python@positions#0", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "positions" } }, { "node_id": "mul@positions#0", "kind": "layer", "op": "torch.mul", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "mul", "name": "torch.mul", "displayName": "torch.mul", "description": "PyTorch mul operation", "icon": "Activity", "category": "tensor", "parameters": [ { "name": "dim", "type": "string", "defaultValue": "", "required": false, "description": "Dimension(s) to apply the operation on" } ] }, "parentId": "positions" } }, { "node_id": "stack@positions#0", "kind": "layer", "op": "torch.stack", "params": { "dim": -1 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "stack", "name": "torch.stack", "displayName": "torch.stack", "description": "PyTorch stack operation", "icon": "Activity", "category": "tensor", "parameters": [ { "name": "dim", "type": "string", "defaultValue": "", "required": false, "description": "Dimension(s) to apply the operation on" } ] }, "parentId": "positions" } }, { "node_id": "flatten@positions#0", "kind": "layer", "op": "Flatten", "params": { "start_dim": -2 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "flatten", "name": "Flatten", "displayName": "Flatten", "description": "Flattens input tensor contiguously", "icon": "Maximize2", "category": "tensor", "parameters": [ { "name": "start_dim", "type": "number", "defaultValue": 1, "required": false, "description": "First dim to flatten", "min": 0 }, { "name": "end_dim", "type": "number", "defaultValue": -1, "required": false, "description": "Last dim to flatten" } ] }, "parentId": "positions" } }, { "node_id": "add@positions#0", "kind": "layer", "op": "torch.add", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "add", "name": "torch.add", "displayName": "torch.add", "description": "PyTorch add operation", "icon": "Activity", "category": "tensor", "parameters": [ { "name": "dim", "type": "string", "defaultValue": "", "required": false, "description": "Dimension(s) to apply the operation on" } ] }, "parentId": "positions" } }, { "node_id": "block.attention", "kind": "module", "op": "Module", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "module", "name": "Module", "displayName": "Module", "description": "Custom module group (nn.Module)", "icon": "Package", "category": "module", "parameters": [ { "name": "name", "type": "string", "defaultValue": "CustomModule", "required": true, "description": "Module class name" }, { "name": "description", "type": "string", "defaultValue": "", "required": false, "description": "Module description" } ] }, "parentId": "block", "moduleInstanceId": "block.attention", "moduleTemplateId": "model::CausalAttention", "isGroup": true, "moduleMembers": [ "block.attention.query", "python@block.attention#0", "block.attention.key", "python@block.attention#1", "block.attention.value", "python@block.attention#2", "block.attention.scores", "block.attention.softmax", "python@block.attention#3", "block.attention.project" ], "moduleLayerCount": 11 } }, { "node_id": "add@block#0", "kind": "layer", "op": "torch.add", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "add", "name": "torch.add", "displayName": "Attention residual", "description": "PyTorch add operation", "icon": "Activity", "category": "tensor", "parameters": [ { "name": "dim", "type": "string", "defaultValue": "", "required": false, "description": "Dimension(s) to apply the operation on" } ] }, "parentId": "block" } }, { "node_id": "block.norm_attention", "kind": "layer", "op": "LayerNorm", "params": { "normalized_shape": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "layernorm", "name": "LayerNorm", "displayName": "LayerNorm", "description": "Layer Normalization", "icon": "Layers", "category": "norm", "parameters": [ { "name": "normalized_shape", "type": "string", "defaultValue": "512", "required": true, "description": "Input shape for normalization (comma-separated for multi-dim, e.g., 512)" }, { "name": "eps", "type": "number", "defaultValue": 0.00001, "required": false, "description": "Value for numerical stability", "min": 0 }, { "name": "elementwise_affine", "type": "boolean", "defaultValue": true, "required": false, "description": "If True, learnable affine parameters" } ] }, "parentId": "block" } }, { "node_id": "block.ffn", "kind": "module", "op": "Module", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "module", "name": "Module", "displayName": "Module", "description": "Custom module group (nn.Module)", "icon": "Package", "category": "module", "parameters": [ { "name": "name", "type": "string", "defaultValue": "CustomModule", "required": true, "description": "Module class name" }, { "name": "description", "type": "string", "defaultValue": "", "required": false, "description": "Module description" } ] }, "parentId": "block", "moduleInstanceId": "block.ffn", "moduleTemplateId": "model::FeedForward", "isGroup": true, "moduleMembers": [ "block.ffn.expand", "block.ffn.relu", "block.ffn.project" ], "moduleLayerCount": 3 } }, { "node_id": "add@block#1", "kind": "layer", "op": "torch.add", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "add", "name": "torch.add", "displayName": "Feed-forward residual", "description": "PyTorch add operation", "icon": "Activity", "category": "tensor", "parameters": [ { "name": "dim", "type": "string", "defaultValue": "", "required": false, "description": "Dimension(s) to apply the operation on" } ] }, "parentId": "block" } }, { "node_id": "block.norm_ffn", "kind": "layer", "op": "LayerNorm", "params": { "normalized_shape": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "layernorm", "name": "LayerNorm", "displayName": "LayerNorm", "description": "Layer Normalization", "icon": "Layers", "category": "norm", "parameters": [ { "name": "normalized_shape", "type": "string", "defaultValue": "512", "required": true, "description": "Input shape for normalization (comma-separated for multi-dim, e.g., 512)" }, { "name": "eps", "type": "number", "defaultValue": 0.00001, "required": false, "description": "Value for numerical stability", "min": 0 }, { "name": "elementwise_affine", "type": "boolean", "defaultValue": true, "required": false, "description": "If True, learnable affine parameters" } ] }, "parentId": "block" } }, { "node_id": "block.attention.query", "kind": "layer", "op": "Linear", "params": { "in_features": 16, "out_features": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] }, "parentId": "block.attention" } }, { "node_id": "python@block.attention#0", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Split query heads · Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "block.attention" } }, { "node_id": "block.attention.key", "kind": "layer", "op": "Linear", "params": { "in_features": 16, "out_features": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] }, "parentId": "block.attention" } }, { "node_id": "python@block.attention#1", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Split key heads · Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "block.attention" } }, { "node_id": "block.attention.value", "kind": "layer", "op": "Linear", "params": { "in_features": 16, "out_features": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] }, "parentId": "block.attention" } }, { "node_id": "python@block.attention#2", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Split value heads · Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "block.attention" } }, { "node_id": "block.attention.scores", "kind": "module", "op": "Module", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "module", "name": "Module", "displayName": "Module", "description": "Custom module group (nn.Module)", "icon": "Package", "category": "module", "parameters": [ { "name": "name", "type": "string", "defaultValue": "CustomModule", "required": true, "description": "Module class name" }, { "name": "description", "type": "string", "defaultValue": "", "required": false, "description": "Module description" } ] }, "parentId": "block.attention", "moduleInstanceId": "block.attention.scores", "moduleTemplateId": "model::CausalScores", "isGroup": true, "moduleMembers": [ "python@block.attention.scores#0", "python@block.attention.scores#1" ], "moduleLayerCount": 2 } }, { "node_id": "block.attention.softmax", "kind": "layer", "op": "Softmax", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "softmax", "name": "Softmax", "displayName": "Softmax", "description": "Softmax over a dimension", "icon": "TrendingUp", "category": "activation", "parameters": [ { "name": "dim", "type": "number", "defaultValue": 1, "required": true, "description": "Dimension along which Softmax is computed" } ] }, "parentId": "block.attention" } }, { "node_id": "python@block.attention#3", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Mix and join heads · Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "block.attention" } }, { "node_id": "block.attention.project", "kind": "layer", "op": "Linear", "params": { "in_features": 16, "out_features": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] }, "parentId": "block.attention" } }, { "node_id": "python@block.attention.scores#0", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Scale attention scores · Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "block.attention.scores" } }, { "node_id": "python@block.attention.scores#1", "kind": "layer", "op": "Python", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "python", "name": "Python", "displayName": "Mask future positions · Python", "description": "A statement from the source file that has no graph equivalent. Its code is kept verbatim and its shape is unknown until the model is traced.", "icon": "Hash", "category": "misc", "parameters": [] }, "parentId": "block.attention.scores" } }, { "node_id": "block.ffn.expand", "kind": "layer", "op": "Linear", "params": { "in_features": 16, "out_features": 64 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] }, "parentId": "block.ffn" } }, { "node_id": "block.ffn.relu", "kind": "layer", "op": "ReLU", "params": {}, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "relu", "name": "ReLU", "displayName": "ReLU", "description": "Rectified Linear Unit", "icon": "TrendingUp", "category": "activation", "parameters": [ { "name": "inplace", "type": "boolean", "defaultValue": false, "required": false, "description": "Can optionally do the operation in-place" } ] }, "parentId": "block.ffn" } }, { "node_id": "block.ffn.project", "kind": "layer", "op": "Linear", "params": { "in_features": 64, "out_features": 16 }, "in_ports": [ "x" ], "out_ports": [ "y" ], "metadata": { "layerType": { "id": "linear", "name": "Linear", "displayName": "Linear", "description": "Fully Connected Layer", "icon": "Minus", "category": "linear", "parameters": [ { "name": "in_features", "type": "number", "defaultValue": 512, "required": true, "description": "Size of input features", "min": 1 }, { "name": "out_features", "type": "number", "defaultValue": 256, "required": true, "description": "Size of output features", "min": 1 }, { "name": "bias", "type": "boolean", "defaultValue": true, "required": false, "description": "If set to False, the layer will not learn an additive bias" } ] }, "parentId": "block.ffn" } } ], "edges": [ { "edge_id": "edge-1", "source_node_id": "python@positions#0", "target_node_id": "mul@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-2", "source_node_id": "python@positions#1", "target_node_id": "mul@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-3", "source_node_id": "mul@positions#0", "target_node_id": "stack@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-4", "source_node_id": "stack@positions#0", "target_node_id": "flatten@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-5", "source_node_id": "flatten@positions#0", "target_node_id": "add@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-6", "source_node_id": "python@block.attention.scores#0", "target_node_id": "python@block.attention.scores#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-7", "source_node_id": "block.attention.query", "target_node_id": "python@block.attention#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-8", "source_node_id": "block.attention.key", "target_node_id": "python@block.attention#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-9", "source_node_id": "block.attention.value", "target_node_id": "python@block.attention#2", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-10", "source_node_id": "python@block.attention#0", "target_node_id": "python@block.attention.scores#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-11", "source_node_id": "python@block.attention#0", "target_node_id": "python@block.attention.scores#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-12", "source_node_id": "python@block.attention#0", "target_node_id": "block.attention.scores", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-13", "source_node_id": "python@block.attention#1", "target_node_id": "python@block.attention.scores#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-14", "source_node_id": "python@block.attention#1", "target_node_id": "python@block.attention.scores#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-15", "source_node_id": "python@block.attention#1", "target_node_id": "block.attention.scores", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-16", "source_node_id": "python@block.attention.scores#1", "target_node_id": "block.attention.softmax", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-17", "source_node_id": "block.attention.scores", "target_node_id": "block.attention.softmax", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-18", "source_node_id": "block.attention.softmax", "target_node_id": "python@block.attention#3", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-19", "source_node_id": "python@block.attention#2", "target_node_id": "python@block.attention#3", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-20", "source_node_id": "python@block.attention#3", "target_node_id": "block.attention.project", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-21", "source_node_id": "block.ffn.expand", "target_node_id": "block.ffn.relu", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-22", "source_node_id": "block.ffn.relu", "target_node_id": "block.ffn.project", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-23", "source_node_id": "block.attention.project", "target_node_id": "add@block#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-24", "source_node_id": "block.attention", "target_node_id": "add@block#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-25", "source_node_id": "add@block#0", "target_node_id": "block.norm_attention", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-26", "source_node_id": "block.norm_attention", "target_node_id": "block.ffn.expand", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-27", "source_node_id": "block.norm_attention", "target_node_id": "block.ffn", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-28", "source_node_id": "block.norm_attention", "target_node_id": "add@block#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-29", "source_node_id": "block.ffn.project", "target_node_id": "add@block#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-30", "source_node_id": "block.ffn", "target_node_id": "add@block#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-31", "source_node_id": "add@block#1", "target_node_id": "block.norm_ffn", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-32", "source_node_id": "input", "target_node_id": "embedding", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-33", "source_node_id": "embedding", "target_node_id": "python@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-34", "source_node_id": "embedding", "target_node_id": "python@positions#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-35", "source_node_id": "embedding", "target_node_id": "add@positions#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-36", "source_node_id": "embedding", "target_node_id": "positions", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-37", "source_node_id": "add@positions#0", "target_node_id": "block.attention.query", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-38", "source_node_id": "add@positions#0", "target_node_id": "python@block.attention#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-39", "source_node_id": "add@positions#0", "target_node_id": "block.attention.key", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-40", "source_node_id": "add@positions#0", "target_node_id": "python@block.attention#1", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-41", "source_node_id": "add@positions#0", "target_node_id": "block.attention.value", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-42", "source_node_id": "add@positions#0", "target_node_id": "python@block.attention#2", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-43", "source_node_id": "add@positions#0", "target_node_id": "python@block.attention#3", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-44", "source_node_id": "add@positions#0", "target_node_id": "add@block#0", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-45", "source_node_id": "positions", "target_node_id": "block", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-46", "source_node_id": "block.norm_ffn", "target_node_id": "vocabulary", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-47", "source_node_id": "block", "target_node_id": "vocabulary", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-48", "source_node_id": "vocabulary", "target_node_id": "probabilities", "source_port": "y", "target_port": "x", "metadata": {} }, { "edge_id": "edge-49", "source_node_id": "probabilities", "target_node_id": "output", "source_port": "y", "target_port": "x", "metadata": {} } ], "inputs": [], "modules": [], "operations": [], "metadata": {} }, "view": { "graph_id": "snapshot", "node_positions": { "input": { "x": 0, "y": 160 }, "embedding": { "x": 340, "y": 160 }, "positions": { "x": 680, "y": 160 }, "block": { "x": 1020, "y": 160 }, "vocabulary": { "x": 1360, "y": 160 }, "probabilities": { "x": 1700, "y": 160 }, "output": { "x": 2040, "y": 160 }, "python@positions#1": { "x": 24, "y": 214 }, "python@positions#0": { "x": 24, "y": 364 }, "mul@positions#0": { "x": 24, "y": 514 }, "stack@positions#0": { "x": 24, "y": 664 }, "flatten@positions#0": { "x": 24, "y": 814 }, "add@positions#0": { "x": 24, "y": 964 }, "block.attention": { "x": 24, "y": 214 }, "add@block#0": { "x": 24, "y": 364 }, "block.norm_attention": { "x": 24, "y": 514 }, "block.ffn": { "x": 24, "y": 664 }, "add@block#1": { "x": 24, "y": 814 }, "block.norm_ffn": { "x": 24, "y": 964 }, "block.attention.query": { "x": 24, "y": 214 }, "python@block.attention#0": { "x": 24, "y": 364 }, "block.attention.key": { "x": 24, "y": 514 }, "python@block.attention#1": { "x": 24, "y": 664 }, "block.attention.value": { "x": 24, "y": 814 }, "python@block.attention#2": { "x": 24, "y": 964 }, "block.attention.scores": { "x": 24, "y": 1114 }, "block.attention.softmax": { "x": 24, "y": 1264 }, "python@block.attention#3": { "x": 24, "y": 1414 }, "block.attention.project": { "x": 24, "y": 1564 }, "python@block.attention.scores#0": { "x": 24, "y": 214 }, "python@block.attention.scores#1": { "x": 24, "y": 364 }, "block.ffn.expand": { "x": 24, "y": 214 }, "block.ffn.relu": { "x": 24, "y": 364 }, "block.ffn.project": { "x": 24, "y": 514 } }, "collapsed_modules": [ "positions", "block", "block.attention", "block.ffn", "block.attention.scores" ], "selection": [], "layout": {} } }, "snapshotId": "snapshot-d177c5aa44d90c8", "collaboration": { "walkthrough": { "title": "One token at a time", "steps": [ { "id": "tokens", "title": "Turn IDs into features", "note": "Each integer token ID indexes one of 32 learned vectors with 16 features. The positions block scales embeddings by √16 and adds sinusoidal position information.", "view": { "camera": { "x": -7.5, "y": 60, "z": 1.25 }, "nodePositions": { "input": { "x": 0, "y": 160 }, "embedding": { "x": 340, "y": 160 }, "positions": { "x": 680, "y": 160 }, "block": { "x": 1020, "y": 160 }, "vocabulary": { "x": 1360, "y": 160 }, "probabilities": { "x": 1700, "y": 160 }, "output": { "x": 2040, "y": 160 }, "python@positions#1": { "x": 24, "y": 214 }, "python@positions#0": { "x": 24, "y": 364 }, "mul@positions#0": { "x": 24, "y": 514 }, "stack@positions#0": { "x": 24, "y": 664 }, "flatten@positions#0": { "x": 24, "y": 814 }, "add@positions#0": { "x": 24, "y": 964 }, "block.attention": { "x": 24, "y": 214 }, "add@block#0": { "x": 24, "y": 364 }, "block.norm_attention": { "x": 24, "y": 514 }, "block.ffn": { "x": 24, "y": 664 }, "add@block#1": { "x": 24, "y": 814 }, "block.norm_ffn": { "x": 24, "y": 964 }, "block.attention.query": { "x": 24, "y": 214 }, "python@block.attention#0": { "x": 24, "y": 364 }, "block.attention.key": { "x": 24, "y": 514 }, "python@block.attention#1": { "x": 24, "y": 664 }, "block.attention.value": { "x": 24, "y": 814 }, "python@block.attention#2": { "x": 24, "y": 964 }, "block.attention.scores": { "x": 24, "y": 1114 }, "block.attention.softmax": { "x": 24, "y": 1264 }, "python@block.attention#3": { "x": 24, "y": 1414 }, "block.attention.project": { "x": 24, "y": 1564 }, "python@block.attention.scores#0": { "x": 24, "y": 214 }, "python@block.attention.scores#1": { "x": 24, "y": 364 }, "block.ffn.expand": { "x": 24, "y": 214 }, "block.ffn.relu": { "x": 24, "y": 364 }, "block.ffn.project": { "x": 24, "y": 514 } }, "expandedModuleIds": [], "selectedNodeIds": [ "embedding" ], "viewport": { "width": 1100, "height": 700 } }, "anchors": [ { "target": { "kind": "node", "nodeId": "embedding" }, "snapshotId": "snapshot-d177c5aa44d90c8", "exportedAt": "2026-09-17T05:43:00.236026+00:00", "modelName": "TinyDecoder", "context": "{\"connections\":[\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"embedding\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"add@positions#0\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"embedding\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"positions\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"embedding\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"python@positions#0\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"embedding\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"python@positions#1\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"input\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"embedding\\\",\\\"targetPort\\\":\\\"x\\\"}\"],\"contents\":[],\"node\":{\"id\":\"embedding\",\"kind\":\"layer\",\"op\":\"Embedding\",\"params\":{\"embedding_dim\":16,\"num_embeddings\":32}}}" } ] }, { "id": "causal", "title": "Restrict attention to the prefix", "note": "Four heads each have four features. Query–key scores are scaled by √4 and future positions are masked before softmax. The recorded rows below show the resulting zero weights.", "view": { "camera": { "x": 308.5, "y": -1044, "z": 1.25 }, "nodePositions": { "input": { "x": 0, "y": 160 }, "embedding": { "x": 340, "y": 160 }, "positions": { "x": 680, "y": 160 }, "block": { "x": 1020, "y": 160 }, "vocabulary": { "x": 1360, "y": 160 }, "probabilities": { "x": 1700, "y": 160 }, "output": { "x": 2040, "y": 160 }, "python@positions#1": { "x": 24, "y": 214 }, "python@positions#0": { "x": 24, "y": 364 }, "mul@positions#0": { "x": 24, "y": 514 }, "stack@positions#0": { "x": 24, "y": 664 }, "flatten@positions#0": { "x": 24, "y": 814 }, "add@positions#0": { "x": 24, "y": 964 }, "block.attention": { "x": 24, "y": 214 }, "add@block#0": { "x": 24, "y": 364 }, "block.norm_attention": { "x": 24, "y": 514 }, "block.ffn": { "x": 24, "y": 664 }, "add@block#1": { "x": 24, "y": 814 }, "block.norm_ffn": { "x": 24, "y": 964 }, "block.attention.query": { "x": 24, "y": 214 }, "python@block.attention#0": { "x": 24, "y": 364 }, "block.attention.key": { "x": 24, "y": 514 }, "python@block.attention#1": { "x": 24, "y": 664 }, "block.attention.value": { "x": 24, "y": 814 }, "python@block.attention#2": { "x": 24, "y": 964 }, "block.attention.scores": { "x": 24, "y": 1114 }, "block.attention.softmax": { "x": 24, "y": 1264 }, "python@block.attention#3": { "x": 24, "y": 1414 }, "block.attention.project": { "x": 24, "y": 1564 }, "python@block.attention.scores#0": { "x": 24, "y": 214 }, "python@block.attention.scores#1": { "x": 24, "y": 364 }, "block.ffn.expand": { "x": 24, "y": 214 }, "block.ffn.relu": { "x": 24, "y": 364 }, "block.ffn.project": { "x": 24, "y": 514 } }, "expandedModuleIds": [ "block", "block.attention" ], "selectedNodeIds": [ "block.attention.softmax" ], "viewport": { "width": 1100, "height": 700 } }, "anchors": [ { "target": { "kind": "node", "nodeId": "block.attention.softmax" }, "snapshotId": "snapshot-d177c5aa44d90c8", "exportedAt": "2026-09-17T05:43:00.236026+00:00", "modelName": "TinyDecoder", "context": "{\"connections\":[\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.attention.scores\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.attention.softmax\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.attention.softmax\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"python@block.attention#3\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"python@block.attention.scores#1\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.attention.softmax\\\",\\\"targetPort\\\":\\\"x\\\"}\"],\"contents\":[],\"node\":{\"id\":\"block.attention.softmax\",\"kind\":\"layer\",\"op\":\"Softmax\",\"params\":{},\"parent\":\"block.attention\"}}" } ] }, { "id": "feed-forward", "title": "Transform each token's features", "note": "A 16 → 64 → 16 ReLU feed-forward network processes each position independently. Residual additions and LayerNorm surround the attention and feed-forward routes.", "view": { "camera": { "x": 434.8461538461538, "y": -590.8251748251748, "z": 0.8746177370030581 }, "nodePositions": { "input": { "x": 0, "y": 160 }, "embedding": { "x": 340, "y": 160 }, "positions": { "x": 680, "y": 160 }, "block": { "x": 1020, "y": 160 }, "vocabulary": { "x": 1360, "y": 160 }, "probabilities": { "x": 1700, "y": 160 }, "output": { "x": 2040, "y": 160 }, "python@positions#1": { "x": 24, "y": 214 }, "python@positions#0": { "x": 24, "y": 364 }, "mul@positions#0": { "x": 24, "y": 514 }, "stack@positions#0": { "x": 24, "y": 664 }, "flatten@positions#0": { "x": 24, "y": 814 }, "add@positions#0": { "x": 24, "y": 964 }, "block.attention": { "x": 24, "y": 214 }, "add@block#0": { "x": 24, "y": 364 }, "block.norm_attention": { "x": 24, "y": 514 }, "block.ffn": { "x": 24, "y": 664 }, "add@block#1": { "x": 24, "y": 814 }, "block.norm_ffn": { "x": 24, "y": 964 }, "block.attention.query": { "x": 24, "y": 214 }, "python@block.attention#0": { "x": 24, "y": 364 }, "block.attention.key": { "x": 24, "y": 514 }, "python@block.attention#1": { "x": 24, "y": 664 }, "block.attention.value": { "x": 24, "y": 814 }, "python@block.attention#2": { "x": 24, "y": 964 }, "block.attention.scores": { "x": 24, "y": 1114 }, "block.attention.softmax": { "x": 24, "y": 1264 }, "python@block.attention#3": { "x": 24, "y": 1414 }, "block.attention.project": { "x": 24, "y": 1564 }, "python@block.attention.scores#0": { "x": 24, "y": 214 }, "python@block.attention.scores#1": { "x": 24, "y": 364 }, "block.ffn.expand": { "x": 24, "y": 214 }, "block.ffn.relu": { "x": 24, "y": 364 }, "block.ffn.project": { "x": 24, "y": 514 } }, "expandedModuleIds": [ "block", "block.ffn" ], "selectedNodeIds": [ "block.ffn" ], "viewport": { "width": 1100, "height": 700 } }, "anchors": [ { "target": { "kind": "node", "nodeId": "block.ffn" }, "snapshotId": "snapshot-d177c5aa44d90c8", "exportedAt": "2026-09-17T05:43:00.236026+00:00", "modelName": "TinyDecoder", "context": "{\"connections\":[\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.ffn\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"add@block#1\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.ffn.expand\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.ffn.relu\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.ffn.project\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"add@block#1\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.ffn.relu\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.ffn.project\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.norm_attention\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.ffn\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.norm_attention\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.ffn.expand\\\",\\\"targetPort\\\":\\\"x\\\"}\"],\"contents\":[\"{\\\"id\\\":\\\"block.ffn.expand\\\",\\\"kind\\\":\\\"layer\\\",\\\"op\\\":\\\"Linear\\\",\\\"params\\\":{\\\"in_features\\\":16,\\\"out_features\\\":64},\\\"parent\\\":\\\"block.ffn\\\"}\",\"{\\\"id\\\":\\\"block.ffn.project\\\",\\\"kind\\\":\\\"layer\\\",\\\"op\\\":\\\"Linear\\\",\\\"params\\\":{\\\"in_features\\\":64,\\\"out_features\\\":16},\\\"parent\\\":\\\"block.ffn\\\"}\",\"{\\\"id\\\":\\\"block.ffn.relu\\\",\\\"kind\\\":\\\"layer\\\",\\\"op\\\":\\\"ReLU\\\",\\\"params\\\":{},\\\"parent\\\":\\\"block.ffn\\\"}\"],\"node\":{\"id\":\"block.ffn\",\"kind\":\"module\",\"op\":\"Module\",\"params\":{},\"parent\":\"block\"}}" } ] }, { "id": "predict", "title": "Form a vocabulary distribution", "note": "The final projection creates 32 logits for every position. Softmax turns them into probabilities that sum to one; training would teach these outputs to predict the next token.", "view": { "camera": { "x": -1367.5, "y": 60, "z": 1.25 }, "nodePositions": { "input": { "x": 0, "y": 160 }, "embedding": { "x": 340, "y": 160 }, "positions": { "x": 680, "y": 160 }, "block": { "x": 1020, "y": 160 }, "vocabulary": { "x": 1360, "y": 160 }, "probabilities": { "x": 1700, "y": 160 }, "output": { "x": 2040, "y": 160 }, "python@positions#1": { "x": 24, "y": 214 }, "python@positions#0": { "x": 24, "y": 364 }, "mul@positions#0": { "x": 24, "y": 514 }, "stack@positions#0": { "x": 24, "y": 664 }, "flatten@positions#0": { "x": 24, "y": 814 }, "add@positions#0": { "x": 24, "y": 964 }, "block.attention": { "x": 24, "y": 214 }, "add@block#0": { "x": 24, "y": 364 }, "block.norm_attention": { "x": 24, "y": 514 }, "block.ffn": { "x": 24, "y": 664 }, "add@block#1": { "x": 24, "y": 814 }, "block.norm_ffn": { "x": 24, "y": 964 }, "block.attention.query": { "x": 24, "y": 214 }, "python@block.attention#0": { "x": 24, "y": 364 }, "block.attention.key": { "x": 24, "y": 514 }, "python@block.attention#1": { "x": 24, "y": 664 }, "block.attention.value": { "x": 24, "y": 814 }, "python@block.attention#2": { "x": 24, "y": 964 }, "block.attention.scores": { "x": 24, "y": 1114 }, "block.attention.softmax": { "x": 24, "y": 1264 }, "python@block.attention#3": { "x": 24, "y": 1414 }, "block.attention.project": { "x": 24, "y": 1564 }, "python@block.attention.scores#0": { "x": 24, "y": 214 }, "python@block.attention.scores#1": { "x": 24, "y": 364 }, "block.ffn.expand": { "x": 24, "y": 214 }, "block.ffn.relu": { "x": 24, "y": 364 }, "block.ffn.project": { "x": 24, "y": 514 } }, "expandedModuleIds": [], "selectedNodeIds": [ "probabilities" ], "viewport": { "width": 1100, "height": 700 } }, "anchors": [ { "target": { "kind": "node", "nodeId": "probabilities" }, "snapshotId": "snapshot-d177c5aa44d90c8", "exportedAt": "2026-09-17T05:43:00.236026+00:00", "modelName": "TinyDecoder", "context": "{\"connections\":[\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"probabilities\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"output\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"vocabulary\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"probabilities\\\",\\\"targetPort\\\":\\\"x\\\"}\"],\"contents\":[],\"node\":{\"id\":\"probabilities\",\"kind\":\"layer\",\"op\":\"Softmax\",\"params\":{}}}" } ] } ] }, "threads": [], "savedViews": [ { "id": "causal", "title": "Restrict attention to the prefix", "note": "Four heads each have four features. Query–key scores are scaled by √4 and future positions are masked before softmax. The recorded rows below show the resulting zero weights.", "view": { "camera": { "x": 308.5, "y": -1044, "z": 1.25 }, "nodePositions": { "input": { "x": 0, "y": 160 }, "embedding": { "x": 340, "y": 160 }, "positions": { "x": 680, "y": 160 }, "block": { "x": 1020, "y": 160 }, "vocabulary": { "x": 1360, "y": 160 }, "probabilities": { "x": 1700, "y": 160 }, "output": { "x": 2040, "y": 160 }, "python@positions#1": { "x": 24, "y": 214 }, "python@positions#0": { "x": 24, "y": 364 }, "mul@positions#0": { "x": 24, "y": 514 }, "stack@positions#0": { "x": 24, "y": 664 }, "flatten@positions#0": { "x": 24, "y": 814 }, "add@positions#0": { "x": 24, "y": 964 }, "block.attention": { "x": 24, "y": 214 }, "add@block#0": { "x": 24, "y": 364 }, "block.norm_attention": { "x": 24, "y": 514 }, "block.ffn": { "x": 24, "y": 664 }, "add@block#1": { "x": 24, "y": 814 }, "block.norm_ffn": { "x": 24, "y": 964 }, "block.attention.query": { "x": 24, "y": 214 }, "python@block.attention#0": { "x": 24, "y": 364 }, "block.attention.key": { "x": 24, "y": 514 }, "python@block.attention#1": { "x": 24, "y": 664 }, "block.attention.value": { "x": 24, "y": 814 }, "python@block.attention#2": { "x": 24, "y": 964 }, "block.attention.scores": { "x": 24, "y": 1114 }, "block.attention.softmax": { "x": 24, "y": 1264 }, "python@block.attention#3": { "x": 24, "y": 1414 }, "block.attention.project": { "x": 24, "y": 1564 }, "python@block.attention.scores#0": { "x": 24, "y": 214 }, "python@block.attention.scores#1": { "x": 24, "y": 364 }, "block.ffn.expand": { "x": 24, "y": 214 }, "block.ffn.relu": { "x": 24, "y": 364 }, "block.ffn.project": { "x": 24, "y": 514 } }, "expandedModuleIds": [ "block", "block.attention" ], "selectedNodeIds": [ "block.attention.softmax" ], "viewport": { "width": 1100, "height": 700 } }, "anchors": [ { "target": { "kind": "node", "nodeId": "block.attention.softmax" }, "snapshotId": "snapshot-d177c5aa44d90c8", "exportedAt": "2026-09-17T05:43:00.236026+00:00", "modelName": "TinyDecoder", "context": "{\"connections\":[\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.attention.scores\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.attention.softmax\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"block.attention.softmax\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"python@block.attention#3\\\",\\\"targetPort\\\":\\\"x\\\"}\",\"{\\\"kind\\\":\\\"edge\\\",\\\"source\\\":\\\"python@block.attention.scores#1\\\",\\\"sourcePort\\\":\\\"y\\\",\\\"target\\\":\\\"block.attention.softmax\\\",\\\"targetPort\\\":\\\"x\\\"}\"],\"contents\":[],\"node\":{\"id\":\"block.attention.softmax\",\"kind\":\"layer\",\"op\":\"Softmax\",\"params\":{},\"parent\":\"block.attention\"}}" } ] } ], "openingViewId": "causal" } }