Limit to gen-12lp

Author: qjia7

js/web/lib/wasm/jsep/webgpu/ops/matmulnbits.ts

@@ -436,7 +436,11 @@ export const createMatMulNBitsBlockSize32ProgramInfo = (
 
 export const matMulNBits = (context: ComputeContext, attributes: MatMulNBitsAttributes): void => {
   validateInputs(context.inputs, attributes);
-  if (attributes.blockSize === 32 && context.adapterInfo.isVendor('intel')) {
+  if (
+    attributes.blockSize === 32 &&
+    context.adapterInfo.isVendor('intel') &&
+    context.adapterInfo.isArchitecture('gen-12lp')
+  ) {
     context.compute(createMatMulNBitsBlockSize32ProgramInfo(context.inputs, attributes));
   } else {
     context.compute(createMatMulNBitsProgramInfo(context.inputs, attributes));

js/web/lib/wasm/jsep/webgpu/types.ts

@@ -15,7 +15,7 @@ export enum GpuDataType {
 }
 export type GpuDataId = number;
 
-export type GpuArchitecture = 'ampere';
+export type GpuArchitecture = 'ampere' | 'gen-12lp';
 export type GpuVendor = 'amd' | 'intel' | 'nvidia';
 export interface AdapterInfo {
   isArchitecture: (architecture: GpuArchitecture) => boolean;