{ "op": "com.microsoft.MatMulBnb4", "cases": [ { "name": "bnb4-nf4-decode-m1-k2048-n2048-b64", "preset": "smoke", "vars": { "M": 1, "K": 2048, "N": 2048, "blockSize": 64 }, "attrs": { "K": 2048, "N": 2048, "block_size": 64, "quant_type": 1 }, "inputs": { "aT": { "shape": [1, 2048], "dtype": "float32", "dist": "normal", "seed": 410, "scale": 0.2 }, "bT": { "shape": [2097152], "dtype": "uint8", "dist": "linearMod", "mod": 256, "step": 7 }, "absmaxT": { "shape": [65536], "dtype": "float32", "dist": "uniform", "seed": 411, "offset": 0.04, "scale": 0.01, "signed": false } }, "outputs": { "yT": { "shape": [1, 2048], "dtype": "float32" } }, "bench": { "metrics": [ { "type": "bandwidth", "value": "numel(shapes.bT) * 4 + (numel(shapes.aT) + numel(shapes.absmaxT) + numel(shapes.yT)) * 4" } ] } }, { "name": "bnb4-nf4-prefill-m128-k2048-n2048-b64", "preset": "smoke", "vars": { "M": 128, "K": 2048, "N": 2048, "blockSize": 64 }, "attrs": { "K": 2048, "N": 2048, "block_size": 64, "quant_type": 1 }, "tunableSpace": { "PORTABLE_TILE_K": [16, 32, 64] }, "inputs": { "aT": { "shape": [128, 2048], "dtype": "float32", "dist": "normal", "seed": 412, "scale": 0.2 }, "bT": { "shape": [2097152], "dtype": "uint8", "dist": "linearMod", "mod": 256, "step": 7 }, "absmaxT": { "shape": [65536], "dtype": "float32", "dist": "uniform", "seed": 413, "offset": 0.04, "scale": 0.01, "signed": false } }, "outputs": { "yT": { "shape": [128, 2048], "dtype": "float32" } }, "bench": { "primary": true, "metrics": [{ "type": "gflops", "value": "2 * dim(shapes.aT, 0) * args.K * args.N" }] } }, { "name": "bnb4-fp4-prefill-m128-k2048-n2048-b64", "preset": "smoke", "vars": { "M": 128, "K": 2048, "N": 2048, "blockSize": 64 }, "attrs": { "K": 2048, "N": 2048, "block_size": 64, "quant_type": 0 }, "tunableSpace": { "PORTABLE_TILE_K": [16, 32, 64] }, "inputs": { "aT": { "shape": [128, 2048], "dtype": "float32", "dist": "normal", "seed": 414, "scale": 0.2 }, "bT": { "shape": [2097152], "dtype": "uint8", "dist": "linearMod", "mod": 256, "step": 7 }, "absmaxT": { "shape": [65536], "dtype": "float32", "dist": "uniform", "seed": 415, "offset": 0.04, "scale": 0.01, "signed": false } }, "outputs": { "yT": { "shape": [128, 2048], "dtype": "float32" } }, "bench": { "metrics": [{ "type": "gflops", "value": "2 * dim(shapes.aT, 0) * args.K * args.N" }] } }, { "name": "bnb4-nf4-prefill-m512-k4096-n4096-b64", "preset": "model", "vars": { "M": 512, "K": 4096, "N": 4096, "blockSize": 64 }, "attrs": { "K": 4096, "N": 4096, "block_size": 64, "quant_type": 1 }, "inputs": { "aT": { "shape": [512, 4096], "dtype": "float32", "dist": "normal", "seed": 416, "scale": 0.2 }, "bT": { "shape": [8388608], "dtype": "uint8", "dist": "linearMod", "mod": 256, "step": 7 }, "absmaxT": { "shape": [262144], "dtype": "float32", "dist": "uniform", "seed": 417, "offset": 0.04, "scale": 0.01, "signed": false } }, "outputs": { "yT": { "shape": [512, 4096], "dtype": "float32" } }, "bench": { "metrics": [{ "type": "gflops", "value": "2 * dim(shapes.aT, 0) * args.K * args.N" }] }, "tunableSpace": { "PORTABLE_TILE_K": [16, 32, 64] } }, { "name": "bnb4-nf4-prefill-m512-k4096-n4096-b64-f16", "preset": "model", "vars": { "M": 512, "K": 4096, "N": 4096, "blockSize": 64 }, "attrs": { "K": 4096, "N": 4096, "block_size": 64, "quant_type": 1 }, "inputs": { "aT": { "shape": [512, 4096], "dtype": "float16", "dist": "normal", "seed": 416, "scale": 0.2 }, "bT": { "shape": [8388608], "dtype": "uint8", "dist": "linearMod", "mod": 256, "step": 7 }, "absmaxT": { "shape": [262144], "dtype": "float16", "dist": "uniform", "seed": 417, "offset": 0.04, "scale": 0.01, "signed": false } }, "outputs": { "yT": { "shape": [512, 4096], "dtype": "float16" } }, "bench": { "metrics": [{ "type": "gflops", "value": "2 * dim(shapes.aT, 0) * args.K * args.N" }] }, "tunableSpace": { "PORTABLE_TILE_K": [16, 32, 64] } } ] }