{ "name": "com.microsoft.QAttention", "id": "_com_microsoft_qattention_webgpu_026c861", "version": 1, "license": "Apache-2.0", "backend": { "type": "webgpu" }, "digest": { "algorithm": "sha256", "files": { "bench.json": "WUARGQXLHdRwRaYeigTM1jvVwWdzu47zbI09h8lRXsc=", "manifest.json": "ulKRNtTBgoA8Wz+aAQFJIL0QWRYUwEjKv9NM7k/uNSg=", "qattention-attend.wgsl.jinja": "UGc6Q7OUoCatgxhKNvEhHNFyJo+Es1skiNgWXQmsuqE=", "qattention-kv-permute.wgsl.jinja": "jUf53tE21TwhtJHLipUtAN8SRpNf802Gv2U1AyzquDs=", "qattention-mask.wgsl.jinja": "C2ro05i/MqgrdCj4ovrE/6Ynfr2GgILZlXN1M7PYQh4=", "qattention-qkv-scatter.wgsl.jinja": "oOYfxtZyxa+p8x3Qa8nBQmR2mfaSuuQqC4bnHtGodGM=", "quant-dp4a-matmul.wgsl.jinja": "aqvfh8ylqkGgD9hpgcF3VaqrfTd9nxFAOajSYefyKIw=", "quant-exact-matrix.wgsl.jinja": "cxPvYjw5Y+zzLhFVmg++2hZHj/4n3My4GCigHAYzZgo=", "test.json": "sYqqYfG5fzcotj12ayetyuHKBoOhjFxY2UXs7liRv5k=" } }, "provenance": { "kernel": { "sha": "6fdf6301e2bbcc2f03bf1eaf493b7ad55ef33afc", "dirty": false } }, "webgpu": { "manifestSpec": "2.1", "variants": { "sgmat_plain_mask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_mask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_mask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_mask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_mask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_mask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_mask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_mask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_nomask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_nomask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_nomask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_nomask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_nomask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_nomask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_plain_nomask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_plain_nomask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_mask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_mask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_mask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_mask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_mask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_mask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_mask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_mask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_nomask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_nomask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_nomask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_nomask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_nomask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_nomask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_present_nomask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_present_nomask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_mask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_mask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_mask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_mask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_mask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_mask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_mask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_mask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_nomask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_nomask_izp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_nomask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_nomask_izp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_nomask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_nomask_noizp_wzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"], "sgmat_past_nomask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-exact-matrix.wgsl.jinja"], "packed_past_nomask_noizp_nowzp": ["qattention-attend.wgsl.jinja", "qattention-kv-permute.wgsl.jinja", "qattention-mask.wgsl.jinja", "qattention-qkv-scatter.wgsl.jinja", "quant-dp4a-matmul.wgsl.jinja"] } } }