| { |
| "domain": "com.microsoft", |
| "name": "MoE", |
| "sinceVersion": 1, |
| "inputs": { |
| "inputT": { "onnx": "input", "dtype": "T" }, |
| "routerT": { "onnx": "router_probs", "dtype": "T", "rank": 2 }, |
| "fc1T": { "onnx": "fc1_experts_weights", "dtype": "T", "rank": 3 }, |
| "fc1BiasT": { "onnx": "fc1_experts_bias", "dtype": "T", "rank": 2, "optional": true }, |
| "fc2T": { "onnx": "fc2_experts_weights", "dtype": "T", "rank": 3 }, |
| "fc2BiasT": { "onnx": "fc2_experts_bias", "dtype": "T", "rank": 2, "optional": true }, |
| "fc3T": { "onnx": "fc3_experts_weights", "dtype": "T", "rank": 3, "optional": true }, |
| "fc3BiasT": { "onnx": "fc3_experts_bias", "dtype": "T", "rank": 2, "optional": true } |
| }, |
| "outputs": { "outputT": { "onnx": "output", "dtype": "T", "rank": "ranks.inputT", "shape": "shapes.inputT" } }, |
| "attributes": { |
| "activation_alpha": { "default": 1 }, |
| "activation_beta": { "default": 0 }, |
| "activation_type": { "default": "relu" }, |
| "k": { "default": 1 }, |
| "normalize_routing_weights": { "default": 0 }, |
| "swiglu_fusion": { "default": 0 }, |
| "use_sparse_mixer": { "default": 0 }, |
| "swiglu_limit": {} |
| }, |
| "attributeConstraints": { |
| "activation_type": { "values": ["relu", "gelu", "silu", "swiglu", "identity"] }, |
| "normalize_routing_weights": { "values": [0, 1] }, |
| "swiglu_fusion": { "values": [0, 1, 2] }, |
| "use_sparse_mixer": { "values": [0] } |
| }, |
| "typeConstraints": { "T": ["float32"] }, |
| "tunables": { |
| "workgroupSize": { "default": 64 }, |
| "decodeLanes": { "default": 32 }, |
| "decodeBlockTarget": { "default": 1024 }, |
| "decodeMinLaneTrips": { "default": 4 }, |
| "groupThreads": { "default": 8 }, |
| "groupRegM": { "default": 4 }, |
| "groupRegN": { "default": 4 }, |
| "groupTileK": { "default": 16 }, |
| "groupRouteWorkgroup": { "default": 256 }, |
| "matrixMinWorkgroups": { "default": 16 } |
| }, |
| "derive": { |
| "deviceWorkgroupCap": "min(device.limits.maxComputeInvocationsPerWorkgroup, device.limits.maxComputeWorkgroupSizeX)", |
| "wave32Adapter": "has(device.adapterInfo, \"subgroupMinSize\") and has(device.adapterInfo, \"subgroupMaxSize\") and device.adapterInfo.subgroupMinSize == 32 and device.adapterInfo.subgroupMaxSize == 32", |
| "canPinSubgroupSize32": "device.features.has(\"subgroups\") and device.features.has(\"subgroup-size-control\") and has(device.adapterInfo, \"subgroupMinSize\") and has(device.adapterInfo, \"subgroupMaxSize\") and device.adapterInfo.subgroupMinSize <= 32 and device.adapterInfo.subgroupMaxSize >= 32", |
| "pinSubgroupSize32": "canPinSubgroupSize32 and not wave32Adapter", |
| "wave32Effective": "wave32Adapter or pinSubgroupSize32", |
| "topK": "attrs.k", |
| "hiddenDim": "dim(shapes.inputT, ranks.inputT - 1)", |
| "tokenCount": "numel(shapes.inputT) / max(1, hiddenDim)", |
| "expertCount": "dim(shapes.routerT, 1)", |
| "activationType": "attrs.activation_type", |
| "isSwiglu": "activationType == \"swiglu\"", |
| "fusionMode": "attrs.swiglu_fusion", |
| "fusionSize": "2 if isSwiglu and fusionMode != 0 else 1", |
| "interSize": "dim(shapes.fc1T, 1) / max(1, fusionSize)", |
| "fusionOk": "(fusionMode == 0 or fusionMode == 1 or fusionMode == 2) if isSwiglu else fusionMode == 0", |
| "fc3ShapeOk": "not present.fc3T or (ranks.fc3T == 3 and dim(shapes.fc3T, 0) == expertCount and dim(shapes.fc3T, 1) == interSize and dim(shapes.fc3T, 2) == hiddenDim)", |
| "fc3PresenceOk": "(present.fc3T if fusionMode == 0 else not present.fc3T) if isSwiglu else (not present.fc3T or activationType == \"silu\")", |
| "fc1BiasOk": "not present.fc1BiasT or (ranks.fc1BiasT == 2 and dim(shapes.fc1BiasT, 0) == expertCount and dim(shapes.fc1BiasT, 1) == dim(shapes.fc1T, 1))", |
| "fc2BiasOk": "not present.fc2BiasT or (ranks.fc2BiasT == 2 and dim(shapes.fc2BiasT, 0) == expertCount and dim(shapes.fc2BiasT, 1) == hiddenDim)", |
| "fc3BiasOk": "not present.fc3BiasT or (present.fc3T and ranks.fc3BiasT == 2 and dim(shapes.fc3BiasT, 0) == expertCount and dim(shapes.fc3BiasT, 1) == interSize)", |
| "shapeOk": "(ranks.inputT == 2 or ranks.inputT == 3) and ranks.outputT == ranks.inputT and sameShape(shapes.inputT, shapes.outputT) and ranks.routerT == 2 and ranks.fc1T == 3 and ranks.fc2T == 3 and hiddenDim > 0 and interSize > 0 and dim(shapes.routerT, 0) == tokenCount and dim(shapes.fc1T, 0) == expertCount and dim(shapes.fc2T, 0) == expertCount and dim(shapes.fc1T, 2) == hiddenDim and dim(shapes.fc1T, 1) % max(1, fusionSize) == 0 and dim(shapes.fc2T, 1) == hiddenDim and dim(shapes.fc2T, 2) == interSize and topK >= 1 and topK <= expertCount", |
| "contractOk": "shapeOk and fusionOk and fc3ShapeOk and fc3PresenceOk and fc1BiasOk and fc2BiasOk and fc3BiasOk", |
| "workgroupSizeOk": "tunables.workgroupSize >= 1 and tunables.workgroupSize <= deviceWorkgroupCap", |
| "hiddenChunkBytes": "topK * interSize * 4", |
| "hiddenChunkFits": "hiddenChunkBytes <= device.limits.maxStorageBufferBindingSize and hiddenChunkBytes <= device.limits.maxBufferSize", |
| "hiddenChunkTokens": "min(tokenCount, max(1, floor(min(device.limits.maxStorageBufferBindingSize, device.limits.maxBufferSize) / max(1, hiddenChunkBytes))))", |
| "hiddenChunkCount": "max(1, ceilDiv(tokenCount, max(1, hiddenChunkTokens)))", |
| "routeScratchBytes": "tokenCount * topK * 4", |
| "routeScratchFits": "routeScratchBytes <= device.limits.maxStorageBufferBindingSize and routeScratchBytes <= device.limits.maxBufferSize", |
| "splitContract": "contractOk and workgroupSizeOk and hiddenChunkFits and routeScratchFits", |
| "groupTileM": "tunables.groupThreads * tunables.groupRegM", |
| "groupTileN": "tunables.groupThreads * tunables.groupRegN", |
| "groupTileKVec": "ceilDiv(tunables.groupTileK, 4)", |
| "groupThreadCount": "tunables.groupThreads * tunables.groupThreads", |
| "groupSlots": "hiddenChunkTokens * topK", |
| "groupMaxTiles": "ceilDiv(groupSlots, max(1, groupTileM)) + expertCount", |
| "groupSharedBytes": "(groupTileM * tunables.groupTileK + 2 * groupTileN * tunables.groupTileK + groupTileM) * 4", |
| "groupedDeviceOk": "groupThreadCount <= device.limits.maxComputeInvocationsPerWorkgroup and tunables.groupThreads <= device.limits.maxComputeWorkgroupSizeX and tunables.groupThreads <= device.limits.maxComputeWorkgroupSizeY and tunables.groupRouteWorkgroup <= deviceWorkgroupCap and groupSharedBytes <= device.limits.maxComputeWorkgroupStorageSize and expertCount * 8 <= device.limits.maxComputeWorkgroupStorageSize and tunables.groupTileK % 4 == 0", |
| "groupedDispatchOk": "groupMaxTiles <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and ceilDiv(interSize, max(1, groupTileN)) <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and ceilDiv(hiddenDim, max(1, groupTileN)) <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535)", |
| "groupedShapeOk": "tokenCount * topK * 4 >= groupTileM * expertCount", |
| "groupSlotOutBytes": "hiddenChunkTokens * topK * hiddenDim * 4", |
| "groupSlotOutFits": "groupSlotOutBytes <= device.limits.maxStorageBufferBindingSize and groupSlotOutBytes <= device.limits.maxBufferSize", |
| "groupedContract": "splitContract and groupedDeviceOk and groupedDispatchOk and groupSlotOutFits and groupedShapeOk", |
| "decodeLanesOk": "tunables.decodeLanes >= 1 and tunables.decodeLanes <= tunables.workgroupSize and tunables.workgroupSize % tunables.decodeLanes == 0", |
| "decodeRows": "max(1, tunables.workgroupSize / max(1, tunables.decodeLanes))", |
| "decodeDeviceOk": "workgroupSizeOk and decodeLanesOk and tunables.decodeLanes <= device.limits.maxComputeWorkgroupSizeX and decodeRows <= device.limits.maxComputeWorkgroupSizeY and tunables.workgroupSize * 8 <= device.limits.maxComputeWorkgroupStorageSize", |
| "decodeDispatchOk": "ceilDiv(interSize, decodeRows) <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and ceilDiv(hiddenDim, decodeRows) <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and hiddenChunkTokens * topK <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535)", |
| "decodeLaneDepth": "min(hiddenDim, interSize) / max(1, tunables.decodeLanes)", |
| "decodeDepthOk": "decodeLaneDepth >= tunables.decodeMinLaneTrips", |
| "splitFfnBlocks": "ceilDiv(tokenCount * topK * interSize, max(1, tunables.workgroupSize))", |
| "decodeOccupancyOk": "splitFfnBlocks <= tunables.decodeBlockTarget", |
| "decodeContract": "splitContract and decodeDeviceOk and decodeDispatchOk and decodeDepthOk and (decodeOccupancyOk or not groupedContract)", |
| "groupedVec4Ok": "hiddenDim % 4 == 0 and interSize % 4 == 0", |
| "hidden": "hiddenDim", |
| "inter": "interSize", |
| "fc1Rows": "dim(shapes.fc1T, 1)", |
| "experts": "expertCount", |
| "tokens": "tokenCount", |
| "activation": "activationType", |
| "swigluFusion": "fusionMode", |
| "hasSwigluLimit": "has(attrs, \"swiglu_limit\")", |
| "swigluLimit": "attrs.swiglu_limit if has(attrs, \"swiglu_limit\") else 0", |
| "hasFc1Bias": "present.fc1BiasT", |
| "hasFc2Bias": "present.fc2BiasT", |
| "hasFc3": "present.fc3T", |
| "hasFc3Bias": "present.fc3BiasT", |
| "workgroupSize": "tunables.workgroupSize", |
| "groupedSgmatSharedBytes": "max(groupTileM * 32, 2 * 4 * 4 * 64) * 4", |
| "groupedSgmatOk": "splitContract and groupSlotOutFits and groupedShapeOk and groupTileM == 32 and groupMaxTiles <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64) <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and ceilDiv(hiddenDim, 64) <= min(device.limits.maxComputeWorkgroupsPerDimension, 65535) and deviceWorkgroupCap >= 128 and tunables.groupRouteWorkgroup >= 1 and tunables.groupRouteWorkgroup <= deviceWorkgroupCap and expertCount * 8 <= device.limits.maxComputeWorkgroupStorageSize and groupedSgmatSharedBytes <= device.limits.maxComputeWorkgroupStorageSize and hiddenDim % 64 == 0 and interSize % (32 if present.fc3T or isSwiglu else 64) == 0" |
| }, |
| "bindings": { |
| "router_probs": { "arg": "routerT", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_expert": { "scratch": "routeExpert", "buffer": "storage", "elementType": "u32" }, |
| "route_mix": { "scratch": "routeMix", "buffer": "storage", "elementType": "f32" }, |
| "params": { |
| "buffer": "uniform", |
| "struct": [{ "name": "normalizeRoutingWeights", "type": "u32", "value": "attrs.normalize_routing_weights" }] |
| }, |
| "input": { "arg": "inputT", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_expert_2": { |
| "scratch": "routeExpert", |
| "name": "route_expert", |
| "buffer": "read-only-storage", |
| "elementType": "u32" |
| }, |
| "fc1_experts_weights": { "arg": "fc1T", "buffer": "read-only-storage", "elementType": "f32" }, |
| "hidden_act": { "scratch": "hiddenAct", "buffer": "storage", "elementType": "f32" }, |
| "params_2": { |
| "name": "params", |
| "buffer": "uniform", |
| "struct": [ |
| { "name": "activationAlpha", "type": "f32", "value": "attrs.activation_alpha" }, |
| { "name": "activationBeta", "type": "f32", "value": "attrs.activation_beta" }, |
| { "name": "tokenOffset", "type": "u32", "value": "repeat.chunk * hiddenChunkTokens" }, |
| { |
| "name": "tokenCount", |
| "type": "u32", |
| "value": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| ] |
| }, |
| "params_3": { |
| "name": "params", |
| "buffer": "uniform", |
| "struct": [ |
| { "name": "activationAlpha", "type": "f32", "value": "attrs.activation_alpha" }, |
| { "name": "activationBeta", "type": "f32", "value": "attrs.activation_beta" }, |
| { "name": "tokenOffset", "type": "u32", "value": "repeat.chunk * hiddenChunkTokens" } |
| ] |
| }, |
| "fc3_experts_weights": { "arg": "fc3T", "buffer": "read-only-storage", "elementType": "f32" }, |
| "fc3_experts_bias": { "arg": "fc3BiasT", "buffer": "read-only-storage", "elementType": "f32" }, |
| "fc1_experts_bias": { "arg": "fc1BiasT", "buffer": "read-only-storage", "elementType": "f32" }, |
| "hidden_act_2": { |
| "scratch": "hiddenAct", |
| "name": "hidden_act", |
| "buffer": "read-only-storage", |
| "elementType": "f32" |
| }, |
| "route_mix_2": { "scratch": "routeMix", "name": "route_mix", "buffer": "read-only-storage", "elementType": "f32" }, |
| "fc2_experts_weights": { "arg": "fc2T", "buffer": "read-only-storage", "elementType": "f32" }, |
| "output": { "arg": "outputT", "buffer": "storage", "elementType": "f32" }, |
| "params_4": { |
| "name": "params", |
| "buffer": "uniform", |
| "struct": [ |
| { "name": "tokenOffset", "type": "u32", "value": "repeat.chunk * hiddenChunkTokens" }, |
| { |
| "name": "tokenCount", |
| "type": "u32", |
| "value": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| ] |
| }, |
| "params_5": { |
| "name": "params", |
| "buffer": "uniform", |
| "struct": [{ "name": "tokenOffset", "type": "u32", "value": "repeat.chunk * hiddenChunkTokens" }] |
| }, |
| "fc2_experts_bias": { "arg": "fc2BiasT", "buffer": "read-only-storage", "elementType": "f32" }, |
| "input_2": { "arg": "inputT", "name": "input", "buffer": "read-only-storage", "elementType": "$groupElem" }, |
| "slot_list_2": { "scratch": "slotList", "name": "slot_list", "buffer": "read-only-storage", "elementType": "u32" }, |
| "tile_meta_2": { "scratch": "tileMeta", "name": "tile_meta", "buffer": "read-only-storage", "elementType": "u32" }, |
| "fc1_experts_weights_2": { |
| "arg": "fc1T", |
| "name": "fc1_experts_weights", |
| "buffer": "read-only-storage", |
| "elementType": "$groupElem" |
| }, |
| "fc3_experts_weights_2": { |
| "arg": "fc3T", |
| "name": "fc3_experts_weights", |
| "buffer": "read-only-storage", |
| "elementType": "$groupElem" |
| }, |
| "hidden_act_3": { |
| "scratch": "hiddenAct", |
| "name": "hidden_act", |
| "buffer": "read-only-storage", |
| "elementType": "$groupElem" |
| }, |
| "fc2_experts_weights_2": { |
| "arg": "fc2T", |
| "name": "fc2_experts_weights", |
| "buffer": "read-only-storage", |
| "elementType": "$groupElem" |
| }, |
| "slot_out": { "scratch": "slotOut", "buffer": "storage", "elementType": "f32" }, |
| "matrix_params": { |
| "name": "params", |
| "buffer": "uniform", |
| "struct": [{ "name": "tokenOffset", "type": "u32", "value": "repeat.chunk * hiddenChunkTokens" }] |
| } |
| }, |
| "variants": [ |
| { |
| "id": "split_routed_fc1plain_fc3none_fc2plain", |
| "priority": 10, |
| "when": ["splitContract", "not present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1plain_fc3none_fc2plain", |
| "priority": 20, |
| "when": ["decodeContract", "not present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1plain_fc3none_fc2bias", |
| "priority": 10, |
| "when": ["splitContract", "not present.fc1BiasT", "not present.fc3T", "present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1plain_fc3none_fc2bias", |
| "priority": 20, |
| "when": ["decodeContract", "not present.fc1BiasT", "not present.fc3T", "present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1plain_fc3plain_fc2plain", |
| "priority": 10, |
| "when": ["splitContract", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1plain_fc3plain_fc2plain", |
| "priority": 20, |
| "when": ["decodeContract", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1plain_fc3plain_fc2bias", |
| "priority": 10, |
| "when": ["splitContract", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1plain_fc3plain_fc2bias", |
| "priority": 20, |
| "when": ["decodeContract", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1plain_fc3biased_fc2plain", |
| "priority": 10, |
| "when": ["splitContract", "not present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1plain_fc3biased_fc2plain", |
| "priority": 20, |
| "when": ["decodeContract", "not present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1plain_fc3biased_fc2bias", |
| "priority": 10, |
| "when": ["splitContract", "not present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1plain_fc3biased_fc2bias", |
| "priority": 20, |
| "when": ["decodeContract", "not present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1bias_fc3none_fc2plain", |
| "priority": 10, |
| "when": ["splitContract", "present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1bias_fc3none_fc2plain", |
| "priority": 20, |
| "when": ["decodeContract", "present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1bias_fc3none_fc2bias", |
| "priority": 10, |
| "when": ["splitContract", "present.fc1BiasT", "not present.fc3T", "present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1bias_fc3none_fc2bias", |
| "priority": 20, |
| "when": ["decodeContract", "present.fc1BiasT", "not present.fc3T", "present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1bias_fc3plain_fc2plain", |
| "priority": 10, |
| "when": ["splitContract", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1bias_fc3plain_fc2plain", |
| "priority": 20, |
| "when": ["decodeContract", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1bias_fc3plain_fc2bias", |
| "priority": 10, |
| "when": ["splitContract", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1bias_fc3plain_fc2bias", |
| "priority": 20, |
| "when": ["decodeContract", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1bias_fc3biased_fc2plain", |
| "priority": 10, |
| "when": ["splitContract", "present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1bias_fc3biased_fc2plain", |
| "priority": 20, |
| "when": ["decodeContract", "present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "split_routed_fc1bias_fc3biased_fc2bias", |
| "priority": 10, |
| "when": ["splitContract", "present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT"], |
| "derive": {}, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-stage.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_2"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK * interSize), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-stage.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_4"], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "gemv_routed_fc1bias_fc3biased_fc2bias", |
| "priority": 20, |
| "when": ["decodeContract", "present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { "decodeLanes": "tunables.decodeLanes" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStage", |
| "shader": "moe-ffn-gemv.wgsl.jinja", |
| "bindings": ["input", "route_expert_2", "fc1_experts_weights", "fc1_experts_bias", "fc3_experts_weights", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { |
| "x": "ceilDiv(interSize, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * topK" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStage", |
| "shader": "moe-output-gemv.wgsl.jinja", |
| "bindings": ["hidden_act_2", "route_expert_2", "route_mix_2", "fc2_experts_weights", "fc2_experts_bias", "output", "params_5"], |
| "dispatch": { |
| "x": "ceilDiv(hiddenDim, decodeRows)", |
| "y": "min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens)" |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1plain_fc3none_fc2plain", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "not present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1plain_fc3none_fc2plain", |
| "priority": 30, |
| "when": ["groupedContract", "not present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1plain_fc3none_fc2bias", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "not present.fc1BiasT", "not present.fc3T", "present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1plain_fc3none_fc2bias", |
| "priority": 30, |
| "when": ["groupedContract", "not present.fc1BiasT", "not present.fc3T", "present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1plain_fc3plain_fc2plain", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1plain_fc3plain_fc2plain", |
| "priority": 30, |
| "when": ["groupedContract", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1plain_fc3plain_fc2bias", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1plain_fc3plain_fc2bias", |
| "priority": 30, |
| "when": ["groupedContract", "not present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1plain_fc3biased_fc2plain", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "not present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1plain_fc3biased_fc2plain", |
| "priority": 30, |
| "when": ["groupedContract", "not present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1plain_fc3biased_fc2bias", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "not present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1plain_fc3biased_fc2bias", |
| "priority": 30, |
| "when": ["groupedContract", "not present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1bias_fc3none_fc2plain", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1bias_fc3none_fc2plain", |
| "priority": 30, |
| "when": ["groupedContract", "present.fc1BiasT", "not present.fc3T", "not present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1bias_fc3none_fc2bias", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "present.fc1BiasT", "not present.fc3T", "present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1bias_fc3none_fc2bias", |
| "priority": 30, |
| "when": ["groupedContract", "present.fc1BiasT", "not present.fc3T", "present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1bias_fc3plain_fc2plain", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1bias_fc3plain_fc2plain", |
| "priority": 30, |
| "when": ["groupedContract", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1bias_fc3plain_fc2bias", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1bias_fc3plain_fc2bias", |
| "priority": 30, |
| "when": ["groupedContract", "present.fc1BiasT", "present.fc3T and not present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1bias_fc3biased_fc2plain", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1bias_fc3biased_fc2plain", |
| "priority": 30, |
| "when": ["groupedContract", "present.fc1BiasT", "present.fc3BiasT", "not present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| }, |
| { |
| "id": "sgmat_grouped_routed_fc1bias_fc3biased_fc2bias", |
| "priority": 32, |
| "when": ["groupedSgmatOk", "present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT", "wave32Effective"], |
| "derive": { "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", "groupElem": "\"f32\"" }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "matrix_params"], |
| "dispatch": { |
| "x": "groupMaxTiles", |
| "y": "ceilDiv(interSize * (2 if present.fc3T or isSwiglu else 1), 64)" |
| }, |
| "derive": { |
| "matrixStage": "\"ffn\"", |
| "matrixActivationAlpha": "attrs.activation_alpha", |
| "matrixActivationBeta": "attrs.activation_beta" |
| } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-grouped-sgmat.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, 64)" }, |
| "derive": { "matrixStage": "\"output\"" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ], |
| "requires": { |
| "features": ["subgroups", "chromium-experimental-subgroup-matrix"], |
| "subgroupMatrixConfigs": [{ "componentType": "f32", "resultComponentType": "f32", "M": 8, "N": 8, "K": 8 }] |
| }, |
| "demoteWhen": ["not isSwiglu and not present.fc3T and groupMaxTiles * ceilDiv(interSize, 64) < tunables.matrixMinWorkgroups"] |
| }, |
| { |
| "id": "grouped_routed_fc1bias_fc3biased_fc2bias", |
| "priority": 30, |
| "when": ["groupedContract", "present.fc1BiasT", "present.fc3BiasT", "present.fc2BiasT"], |
| "derive": { |
| "groupTileK": "tunables.groupTileK", |
| "groupThreads": "tunables.groupThreads", |
| "regM": "tunables.groupRegM", |
| "regN": "tunables.groupRegN", |
| "groupRouteWorkgroup": "tunables.groupRouteWorkgroup", |
| "groupVec4": "groupedVec4Ok", |
| "groupElem": "\"vec4<f32>\" if groupedVec4Ok else \"f32\"" |
| }, |
| "intermediates": [ |
| { "id": "routeExpert", "dtype": "uint32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "routeMix", "dtype": "float32", "shape": "[max(1, tokenCount * topK)]" }, |
| { "id": "hiddenAct", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * interSize)]" }, |
| { "id": "slotList", "dtype": "uint32", "shape": "[max(1, groupSlots)]" }, |
| { "id": "tileMeta", "dtype": "uint32", "shape": "[1 + 3 * groupMaxTiles]" }, |
| { "id": "slotOut", "dtype": "float32", "shape": "[max(1, hiddenChunkTokens * topK * hiddenDim)]" } |
| ], |
| "passes": [ |
| { |
| "id": "route_stage", |
| "name": "MoE.RouteStage", |
| "shader": "moe-route-stage.wgsl.jinja", |
| "bindings": ["router_probs", "route_expert", "route_mix", "params"], |
| "dispatch": { |
| "x": "min(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((tokenCount), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "token_chunks", |
| "repeat": { "count": "hiddenChunkCount", "index": "chunk" }, |
| "passes": [ |
| { |
| "id": "group_stage", |
| "name": "MoE.GroupStage", |
| "shader": "expert-group-slots.wgsl.jinja", |
| "bindings": [ |
| "route_expert_2", |
| { "scratch": "slotList", "name": "slot_list", "elementType": "u32" }, |
| { "scratch": "tileMeta", "name": "tile_meta", "elementType": "u32" }, |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "y": "ceilDiv(ceilDiv((groupRouteWorkgroup), (groupRouteWorkgroup)), 65535)", |
| "z": 1 |
| } |
| }, |
| { |
| "id": "ffn_stage", |
| "name": "MoE.FfnStageGrouped", |
| "shader": "moe-ffn-grouped.wgsl.jinja", |
| "bindings": ["input_2", "slot_list_2", "tile_meta_2", "fc1_experts_weights_2", "fc1_experts_bias", "fc3_experts_weights_2", "fc3_experts_bias", "hidden_act", "params_3"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(interSize, groupTileN)" } |
| }, |
| { |
| "id": "output_stage", |
| "name": "MoE.OutputStageGrouped", |
| "shader": "moe-output-grouped.wgsl.jinja", |
| "bindings": ["hidden_act_3", "slot_list_2", "tile_meta_2", "fc2_experts_weights_2", "fc2_experts_bias", "slot_out"], |
| "dispatch": { "x": "groupMaxTiles", "y": "ceilDiv(hiddenDim, groupTileN)" } |
| }, |
| { |
| "id": "mix_stage", |
| "name": "MoE.MixStage", |
| "shader": "moe-mix-stage.wgsl.jinja", |
| "bindings": [ |
| { "scratch": "slotOut", "name": "slot_out", "buffer": "read-only-storage", "elementType": "f32" }, |
| "route_mix_2", |
| "output", |
| "params_4" |
| ], |
| "dispatch": { |
| "x": "min(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "y": "ceilDiv(ceilDiv((min(hiddenChunkTokens, tokenCount - repeat.chunk * hiddenChunkTokens) * hiddenDim), (workgroupSize)), 65535)", |
| "z": 1 |
| } |
| } |
| ] |
| } |
| ] |
| } |
| ] |
| } |
|
|
