diff --git a/.buildkite/kernel_microbenchmarks/all_gather_matmul/w16a16.yml b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w16a16.yml new file mode 100644 index 000000000..478a0bc14 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w16a16.yml @@ -0,0 +1,45 @@ +# all-gather-matmul-w16a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for all-gather-matmul-w16a16" + key: "all-gather-matmul-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for all-gather-matmul-w16a16" + key: "record_all-gather-matmul-w16a16_CorrectnessTest" + depends_on: "all-gather-matmul-w16a16_CorrectnessTest" + env: + CI_TARGET: "all-gather-matmul-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w16a16_CorrectnessTest + + - label: "Performance tests for all-gather-matmul-w16a16" + key: "all-gather-matmul-w16a16_PerformanceTest" + depends_on: "record_all-gather-matmul-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for all-gather-matmul-w16a16" + key: "record_all-gather-matmul-w16a16_PerformanceTest" + depends_on: "all-gather-matmul-w16a16_PerformanceTest" + env: + CI_TARGET: "all-gather-matmul-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a16.yml b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a16.yml new file mode 100644 index 000000000..1a8b32847 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a16.yml @@ -0,0 +1,45 @@ +# all-gather-matmul-w4a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for all-gather-matmul-w4a16" + key: "all-gather-matmul-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for all-gather-matmul-w4a16" + key: "record_all-gather-matmul-w4a16_CorrectnessTest" + depends_on: "all-gather-matmul-w4a16_CorrectnessTest" + env: + CI_TARGET: "all-gather-matmul-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w4a16_CorrectnessTest + + - label: "Performance tests for all-gather-matmul-w4a16" + key: "all-gather-matmul-w4a16_PerformanceTest" + depends_on: "record_all-gather-matmul-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for all-gather-matmul-w4a16" + key: "record_all-gather-matmul-w4a16_PerformanceTest" + depends_on: "all-gather-matmul-w4a16_PerformanceTest" + env: + CI_TARGET: "all-gather-matmul-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a4.yml b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a4.yml new file mode 100644 index 000000000..da764ad44 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a4.yml @@ -0,0 +1,45 @@ +# all-gather-matmul-w4a4 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for all-gather-matmul-w4a4" + key: "all-gather-matmul-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for all-gather-matmul-w4a4" + key: "record_all-gather-matmul-w4a4_CorrectnessTest" + depends_on: "all-gather-matmul-w4a4_CorrectnessTest" + env: + CI_TARGET: "all-gather-matmul-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w4a4_CorrectnessTest + + - label: "Performance tests for all-gather-matmul-w4a4" + key: "all-gather-matmul-w4a4_PerformanceTest" + depends_on: "record_all-gather-matmul-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for all-gather-matmul-w4a4" + key: "record_all-gather-matmul-w4a4_PerformanceTest" + depends_on: "all-gather-matmul-w4a4_PerformanceTest" + env: + CI_TARGET: "all-gather-matmul-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a8.yml b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a8.yml new file mode 100644 index 000000000..520dde456 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w4a8.yml @@ -0,0 +1,45 @@ +# all-gather-matmul-w4a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for all-gather-matmul-w4a8" + key: "all-gather-matmul-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for all-gather-matmul-w4a8" + key: "record_all-gather-matmul-w4a8_CorrectnessTest" + depends_on: "all-gather-matmul-w4a8_CorrectnessTest" + env: + CI_TARGET: "all-gather-matmul-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w4a8_CorrectnessTest + + - label: "Performance tests for all-gather-matmul-w4a8" + key: "all-gather-matmul-w4a8_PerformanceTest" + depends_on: "record_all-gather-matmul-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for all-gather-matmul-w4a8" + key: "record_all-gather-matmul-w4a8_PerformanceTest" + depends_on: "all-gather-matmul-w4a8_PerformanceTest" + env: + CI_TARGET: "all-gather-matmul-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/all_gather_matmul/w8a16.yml b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w8a16.yml new file mode 100644 index 000000000..c8982d8d4 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w8a16.yml @@ -0,0 +1,45 @@ +# all-gather-matmul-w8a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for all-gather-matmul-w8a16" + key: "all-gather-matmul-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for all-gather-matmul-w8a16" + key: "record_all-gather-matmul-w8a16_CorrectnessTest" + depends_on: "all-gather-matmul-w8a16_CorrectnessTest" + env: + CI_TARGET: "all-gather-matmul-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w8a16_CorrectnessTest + + - label: "Performance tests for all-gather-matmul-w8a16" + key: "all-gather-matmul-w8a16_PerformanceTest" + depends_on: "record_all-gather-matmul-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for all-gather-matmul-w8a16" + key: "record_all-gather-matmul-w8a16_PerformanceTest" + depends_on: "all-gather-matmul-w8a16_PerformanceTest" + env: + CI_TARGET: "all-gather-matmul-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/all_gather_matmul/w8a8.yml b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w8a8.yml new file mode 100644 index 000000000..9129b6962 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/all_gather_matmul/w8a8.yml @@ -0,0 +1,45 @@ +# all-gather-matmul-w8a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for all-gather-matmul-w8a8" + key: "all-gather-matmul-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for all-gather-matmul-w8a8" + key: "record_all-gather-matmul-w8a8_CorrectnessTest" + depends_on: "all-gather-matmul-w8a8_CorrectnessTest" + env: + CI_TARGET: "all-gather-matmul-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w8a8_CorrectnessTest + + - label: "Performance tests for all-gather-matmul-w8a8" + key: "all-gather-matmul-w8a8_PerformanceTest" + depends_on: "record_all-gather-matmul-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "all-gather-matmul-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for all-gather-matmul-w8a8" + key: "record_all-gather-matmul-w8a8_PerformanceTest" + depends_on: "all-gather-matmul-w8a8_PerformanceTest" + env: + CI_TARGET: "all-gather-matmul-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh all-gather-matmul-w8a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w16a16.yml b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w16a16.yml new file mode 100644 index 000000000..69f471de1 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w16a16.yml @@ -0,0 +1,46 @@ +# attention_kernels-w16a16 +# kernel support matrix microbenchmarks +# For attention kernels, W[x]A[y] denotes KV cache as W, A as compute, and x, y as bit precision-w16a16 +steps: + - label: "Correctness tests for attention kernels-w16a16" + key: "attention_kernels-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for attention kernels-w16a16" + key: "record_attention_kernels-w16a16_CorrectnessTest" + depends_on: "attention_kernels-w16a16_CorrectnessTest" + env: + CI_TARGET: "attention_kernels-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w16a16_CorrectnessTest + + - label: "Performance tests for attention kernels-w16a16" + key: "attention_kernels-w16a16_PerformanceTest" + depends_on: "record_attention_kernels-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for attention kernels-w16a16" + key: "record_attention_kernels-w16a16_PerformanceTest" + depends_on: "attention_kernels-w16a16_PerformanceTest" + env: + CI_TARGET: "attention_kernels-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a16.yml b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a16.yml new file mode 100644 index 000000000..43f79ebbd --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a16.yml @@ -0,0 +1,46 @@ +# attention_kernels-w4a16 +# kernel support matrix microbenchmarks +# For attention kernels, W[x]A[y] denotes KV cache as W, A as compute, and x, y as bit precision-w4a16 +steps: + - label: "Correctness tests for attention kernels-w4a16" + key: "attention_kernels-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for attention kernels-w4a16" + key: "record_attention_kernels-w4a16_CorrectnessTest" + depends_on: "attention_kernels-w4a16_CorrectnessTest" + env: + CI_TARGET: "attention_kernels-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w4a16_CorrectnessTest + + - label: "Performance tests for attention kernels-w4a16" + key: "attention_kernels-w4a16_PerformanceTest" + depends_on: "record_attention_kernels-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for attention kernels-w4a16" + key: "record_attention_kernels-w4a16_PerformanceTest" + depends_on: "attention_kernels-w4a16_PerformanceTest" + env: + CI_TARGET: "attention_kernels-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a4.yml b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a4.yml new file mode 100644 index 000000000..7bf43b97b --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a4.yml @@ -0,0 +1,46 @@ +# attention_kernels-w4a4 +# kernel support matrix microbenchmarks +# For attention kernels, W[x]A[y] denotes KV cache as W, A as compute, and x, y as bit precision-w4a4 +steps: + - label: "Correctness tests for attention kernels-w4a4" + key: "attention_kernels-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for attention kernels-w4a4" + key: "record_attention_kernels-w4a4_CorrectnessTest" + depends_on: "attention_kernels-w4a4_CorrectnessTest" + env: + CI_TARGET: "attention_kernels-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w4a4_CorrectnessTest + + - label: "Performance tests for attention kernels-w4a4" + key: "attention_kernels-w4a4_PerformanceTest" + depends_on: "record_attention_kernels-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for attention kernels-w4a4" + key: "record_attention_kernels-w4a4_PerformanceTest" + depends_on: "attention_kernels-w4a4_PerformanceTest" + env: + CI_TARGET: "attention_kernels-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a8.yml b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a8.yml new file mode 100644 index 000000000..e930b238f --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w4a8.yml @@ -0,0 +1,46 @@ +# attention_kernels-w4a8 +# kernel support matrix microbenchmarks +# For attention kernels, W[x]A[y] denotes KV cache as W, A as compute, and x, y as bit precision-w4a8 +steps: + - label: "Correctness tests for attention kernels-w4a8" + key: "attention_kernels-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for attention kernels-w4a8" + key: "record_attention_kernels-w4a8_CorrectnessTest" + depends_on: "attention_kernels-w4a8_CorrectnessTest" + env: + CI_TARGET: "attention_kernels-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w4a8_CorrectnessTest + + - label: "Performance tests for attention kernels-w4a8" + key: "attention_kernels-w4a8_PerformanceTest" + depends_on: "record_attention_kernels-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for attention kernels-w4a8" + key: "record_attention_kernels-w4a8_PerformanceTest" + depends_on: "attention_kernels-w4a8_PerformanceTest" + env: + CI_TARGET: "attention_kernels-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w8a16.yml b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w8a16.yml new file mode 100644 index 000000000..9e69303e8 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w8a16.yml @@ -0,0 +1,46 @@ +# attention_kernels-w8a16 +# kernel support matrix microbenchmarks +# For attention kernels, W[x]A[y] denotes KV cache as W, A as compute, and x, y as bit precision-w8a16 +steps: + - label: "Correctness tests for attention kernels-w8a16" + key: "attention_kernels-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for attention kernels-w8a16" + key: "record_attention_kernels-w8a16_CorrectnessTest" + depends_on: "attention_kernels-w8a16_CorrectnessTest" + env: + CI_TARGET: "attention_kernels-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w8a16_CorrectnessTest + + - label: "Performance tests for attention kernels-w8a16" + key: "attention_kernels-w8a16_PerformanceTest" + depends_on: "record_attention_kernels-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for attention kernels-w8a16" + key: "record_attention_kernels-w8a16_PerformanceTest" + depends_on: "attention_kernels-w8a16_PerformanceTest" + env: + CI_TARGET: "attention_kernels-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w8a8.yml b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w8a8.yml new file mode 100644 index 000000000..c9e8aef64 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/for_attention_kernels_KV_cache/w8a8.yml @@ -0,0 +1,46 @@ +# attention_kernels-w8a8 +# kernel support matrix microbenchmarks +# For attention kernels, W[x]A[y] denotes KV cache as W, A as compute, and x, y as bit precision-w8a8 +steps: + - label: "Correctness tests for attention kernels-w8a8" + key: "attention_kernels-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for attention kernels-w8a8" + key: "record_attention_kernels-w8a8_CorrectnessTest" + depends_on: "attention_kernels-w8a8_CorrectnessTest" + env: + CI_TARGET: "attention_kernels-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w8a8_CorrectnessTest + + - label: "Performance tests for attention kernels-w8a8" + key: "attention_kernels-w8a8_PerformanceTest" + depends_on: "record_attention_kernels-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "attention_kernels-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for attention kernels-w8a8" + key: "record_attention_kernels-w8a8_PerformanceTest" + depends_on: "attention_kernels-w8a8_PerformanceTest" + env: + CI_TARGET: "attention_kernels-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh attention_kernels-w8a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/fused_moe/w16a16.yml b/.buildkite/kernel_microbenchmarks/fused_moe/w16a16.yml new file mode 100644 index 000000000..278489120 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/fused_moe/w16a16.yml @@ -0,0 +1,45 @@ +# fused moe-w16a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for fused moe-w16a16" + key: "fused_moe-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for fused moe w16a16" + key: "record_fused_moe-w16a16_CorrectnessTest" + depends_on: "fused_moe-w16a16_CorrectnessTest" + env: + CI_TARGET: "fused moe-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w16a16_CorrectnessTest + + - label: "Performance tests for fused moe-w16a16" + key: "fused_moe-w16a16_PerformanceTest" + depends_on: "record_fused_moe-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for fused moe-w16a16" + key: "record_fused_moe-w16a16_PerformanceTest" + depends_on: "fused_moe-w16a16_PerformanceTest" + env: + CI_TARGET: "fused moe-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/fused_moe/w4a16.yml b/.buildkite/kernel_microbenchmarks/fused_moe/w4a16.yml new file mode 100644 index 000000000..c3506cd37 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/fused_moe/w4a16.yml @@ -0,0 +1,45 @@ +# fused moe-w4a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for fused moe-w4a16" + key: "fused_moe-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for fused moe w4a16" + key: "record_fused_moe-w4a16_CorrectnessTest" + depends_on: "fused_moe-w4a16_CorrectnessTest" + env: + CI_TARGET: "fused moe-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w4a16_CorrectnessTest + + - label: "Performance tests for fused moe-w4a16" + key: "fused_moe-w4a16_PerformanceTest" + depends_on: "record_fused_moe-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for fused moe-w4a16" + key: "record_fused_moe-w4a16_PerformanceTest" + depends_on: "fused_moe-w4a16_PerformanceTest" + env: + CI_TARGET: "fused moe-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/fused_moe/w4a4.yml b/.buildkite/kernel_microbenchmarks/fused_moe/w4a4.yml new file mode 100644 index 000000000..4f44597c3 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/fused_moe/w4a4.yml @@ -0,0 +1,45 @@ +# fused moe-w4a4 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for fused moe-w4a4" + key: "fused_moe-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for fused moe w4a4" + key: "record_fused_moe-w4a4_CorrectnessTest" + depends_on: "fused_moe-w4a4_CorrectnessTest" + env: + CI_TARGET: "fused moe-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w4a4_CorrectnessTest + + - label: "Performance tests for fused moe-w4a4" + key: "fused_moe-w4a4_PerformanceTest" + depends_on: "record_fused_moe-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for fused moe-w4a4" + key: "record_fused_moe-w4a4_PerformanceTest" + depends_on: "fused_moe-w4a4_PerformanceTest" + env: + CI_TARGET: "fused moe-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/fused_moe/w4a8.yml b/.buildkite/kernel_microbenchmarks/fused_moe/w4a8.yml new file mode 100644 index 000000000..2c14d3daa --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/fused_moe/w4a8.yml @@ -0,0 +1,45 @@ +# fused moe-w4a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for fused moe-w4a8" + key: "fused_moe-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for fused moe w4a8" + key: "record_fused_moe-w4a8_CorrectnessTest" + depends_on: "fused_moe-w4a8_CorrectnessTest" + env: + CI_TARGET: "fused moe-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w4a8_CorrectnessTest + + - label: "Performance tests for fused moe-w4a8" + key: "fused_moe-w4a8_PerformanceTest" + depends_on: "record_fused_moe-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for fused moe-w4a8" + key: "record_fused_moe-w4a8_PerformanceTest" + depends_on: "fused_moe-w4a8_PerformanceTest" + env: + CI_TARGET: "fused moe-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/fused_moe/w8a16.yml b/.buildkite/kernel_microbenchmarks/fused_moe/w8a16.yml new file mode 100644 index 000000000..b9d1d51ff --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/fused_moe/w8a16.yml @@ -0,0 +1,45 @@ +# fused moe-w8a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for fused moe-w8a16" + key: "fused_moe-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for fused moe w8a16" + key: "record_fused_moe-w8a16_CorrectnessTest" + depends_on: "fused_moe-w8a16_CorrectnessTest" + env: + CI_TARGET: "fused moe-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w8a16_CorrectnessTest + + - label: "Performance tests for fused moe-w8a16" + key: "fused_moe-w8a16_PerformanceTest" + depends_on: "record_fused_moe-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for fused moe-w8a16" + key: "record_fused_moe-w8a16_PerformanceTest" + depends_on: "fused_moe-w8a16_PerformanceTest" + env: + CI_TARGET: "fused moe-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/fused_moe/w8a8.yml b/.buildkite/kernel_microbenchmarks/fused_moe/w8a8.yml new file mode 100644 index 000000000..dda161b7e --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/fused_moe/w8a8.yml @@ -0,0 +1,45 @@ +# fused moe-w8a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for fused moe-w8a8" + key: "fused_moe-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for fused moe w8a8" + key: "record_fused_moe-w8a8_CorrectnessTest" + depends_on: "fused_moe-w8a8_CorrectnessTest" + env: + CI_TARGET: "fused moe-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w8a8_CorrectnessTest + + - label: "Performance tests for fused moe-w8a8" + key: "fused_moe-w8a8_PerformanceTest" + depends_on: "record_fused_moe-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "fused_moe-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for fused moe-w8a8" + key: "record_fused_moe-w8a8_PerformanceTest" + depends_on: "fused_moe-w8a8_PerformanceTest" + env: + CI_TARGET: "fused moe-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh fused_moe-w8a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w16a16.yml b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w16a16.yml new file mode 100644 index 000000000..66bba1e28 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w16a16.yml @@ -0,0 +1,45 @@ +# generic ragged paged attention v3-w16a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for generic ragged paged attention v3-w16a16" + key: "generic_ragged_paged_attention_v3-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for generic ragged paged attention v3 w16a16" + key: "record_generic_ragged_paged_attention_v3-w16a16_CorrectnessTest" + depends_on: "generic_ragged_paged_attention_v3-w16a16_CorrectnessTest" + env: + CI_TARGET: "generic ragged paged attention v3-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w16a16_CorrectnessTest + + - label: "Performance tests for generic ragged paged attention v3-w16a16" + key: "generic_ragged_paged_attention_v3-w16a16_PerformanceTest" + depends_on: "record_generic_ragged_paged_attention_v3-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for generic ragged paged attention v3-w16a16" + key: "record_generic_ragged_paged_attention_v3-w16a16_PerformanceTest" + depends_on: "generic_ragged_paged_attention_v3-w16a16_PerformanceTest" + env: + CI_TARGET: "generic ragged paged attention v3-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a16.yml b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a16.yml new file mode 100644 index 000000000..8194672ad --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a16.yml @@ -0,0 +1,45 @@ +# generic ragged paged attention v3-w4a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for generic ragged paged attention v3-w4a16" + key: "generic_ragged_paged_attention_v3-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for generic ragged paged attention v3 w4a16" + key: "record_generic_ragged_paged_attention_v3-w4a16_CorrectnessTest" + depends_on: "generic_ragged_paged_attention_v3-w4a16_CorrectnessTest" + env: + CI_TARGET: "generic ragged paged attention v3-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w4a16_CorrectnessTest + + - label: "Performance tests for generic ragged paged attention v3-w4a16" + key: "generic_ragged_paged_attention_v3-w4a16_PerformanceTest" + depends_on: "record_generic_ragged_paged_attention_v3-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for generic ragged paged attention v3-w4a16" + key: "record_generic_ragged_paged_attention_v3-w4a16_PerformanceTest" + depends_on: "generic_ragged_paged_attention_v3-w4a16_PerformanceTest" + env: + CI_TARGET: "generic ragged paged attention v3-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a4.yml b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a4.yml new file mode 100644 index 000000000..4debaf7d4 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a4.yml @@ -0,0 +1,45 @@ +# generic ragged paged attention v3-w4a4 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for generic ragged paged attention v3-w4a4" + key: "generic_ragged_paged_attention_v3-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for generic ragged paged attention v3 w4a4" + key: "record_generic_ragged_paged_attention_v3-w4a4_CorrectnessTest" + depends_on: "generic_ragged_paged_attention_v3-w4a4_CorrectnessTest" + env: + CI_TARGET: "generic ragged paged attention v3-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w4a4_CorrectnessTest + + - label: "Performance tests for generic ragged paged attention v3-w4a4" + key: "generic_ragged_paged_attention_v3-w4a4_PerformanceTest" + depends_on: "record_generic_ragged_paged_attention_v3-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for generic ragged paged attention v3-w4a4" + key: "record_generic_ragged_paged_attention_v3-w4a4_PerformanceTest" + depends_on: "generic_ragged_paged_attention_v3-w4a4_PerformanceTest" + env: + CI_TARGET: "generic ragged paged attention v3-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a8.yml b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a8.yml new file mode 100644 index 000000000..c46503d13 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w4a8.yml @@ -0,0 +1,45 @@ +# generic ragged paged attention v3-w4a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for generic ragged paged attention v3-w4a8" + key: "generic_ragged_paged_attention_v3-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for generic ragged paged attention v3 w4a8" + key: "record_generic_ragged_paged_attention_v3-w4a8_CorrectnessTest" + depends_on: "generic_ragged_paged_attention_v3-w4a8_CorrectnessTest" + env: + CI_TARGET: "generic ragged paged attention v3-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w4a8_CorrectnessTest + + - label: "Performance tests for generic ragged paged attention v3-w4a8" + key: "generic_ragged_paged_attention_v3-w4a8_PerformanceTest" + depends_on: "record_generic_ragged_paged_attention_v3-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for generic ragged paged attention v3-w4a8" + key: "record_generic_ragged_paged_attention_v3-w4a8_PerformanceTest" + depends_on: "generic_ragged_paged_attention_v3-w4a8_PerformanceTest" + env: + CI_TARGET: "generic ragged paged attention v3-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w8a16.yml b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w8a16.yml new file mode 100644 index 000000000..0dad69212 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w8a16.yml @@ -0,0 +1,45 @@ +# generic ragged paged attention v3-w8a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for generic ragged paged attention v3-w8a16" + key: "generic_ragged_paged_attention_v3-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for generic ragged paged attention v3 w8a16" + key: "record_generic_ragged_paged_attention_v3-w8a16_CorrectnessTest" + depends_on: "generic_ragged_paged_attention_v3-w8a16_CorrectnessTest" + env: + CI_TARGET: "generic ragged paged attention v3-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w8a16_CorrectnessTest + + - label: "Performance tests for generic ragged paged attention v3-w8a16" + key: "generic_ragged_paged_attention_v3-w8a16_PerformanceTest" + depends_on: "record_generic_ragged_paged_attention_v3-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for generic ragged paged attention v3-w8a16" + key: "record_generic_ragged_paged_attention_v3-w8a16_PerformanceTest" + depends_on: "generic_ragged_paged_attention_v3-w8a16_PerformanceTest" + env: + CI_TARGET: "generic ragged paged attention v3-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w8a8.yml b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w8a8.yml new file mode 100644 index 000000000..810182edc --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/generic_ragged_paged_attention_v3/w8a8.yml @@ -0,0 +1,45 @@ +# generic ragged paged attention v3-w8a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for generic ragged paged attention v3-w8a8" + key: "generic_ragged_paged_attention_v3-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for generic ragged paged attention v3 w8a8" + key: "record_generic_ragged_paged_attention_v3-w8a8_CorrectnessTest" + depends_on: "generic_ragged_paged_attention_v3-w8a8_CorrectnessTest" + env: + CI_TARGET: "generic ragged paged attention v3-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w8a8_CorrectnessTest + + - label: "Performance tests for generic ragged paged attention v3-w8a8" + key: "generic_ragged_paged_attention_v3-w8a8_PerformanceTest" + depends_on: "record_generic_ragged_paged_attention_v3-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "generic_ragged_paged_attention_v3-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for generic ragged paged attention v3-w8a8" + key: "record_generic_ragged_paged_attention_v3-w8a8_PerformanceTest" + depends_on: "generic_ragged_paged_attention_v3-w8a8_PerformanceTest" + env: + CI_TARGET: "generic ragged paged attention v3-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh generic_ragged_paged_attention_v3-w8a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/gmm/w16a16.yml b/.buildkite/kernel_microbenchmarks/gmm/w16a16.yml new file mode 100644 index 000000000..4ba87dbb9 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/gmm/w16a16.yml @@ -0,0 +1,45 @@ +# gmm-w16a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for gmm-w16a16" + key: "gmm-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for gmm-w16a16" + key: "record_gmm-w16a16_CorrectnessTest" + depends_on: "gmm-w16a16_CorrectnessTest" + env: + CI_TARGET: "gmm-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w16a16_CorrectnessTest + + - label: "Performance tests for gmm-w16a16" + key: "gmm-w16a16_PerformanceTest" + depends_on: "record_gmm-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for gmm-w16a16" + key: "record_gmm-w16a16_PerformanceTest" + depends_on: "gmm-w16a16_PerformanceTest" + env: + CI_TARGET: "gmm-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/gmm/w4a16.yml b/.buildkite/kernel_microbenchmarks/gmm/w4a16.yml new file mode 100644 index 000000000..40dc81973 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/gmm/w4a16.yml @@ -0,0 +1,45 @@ +# gmm-w4a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for gmm-w4a16" + key: "gmm-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for gmm-w4a16" + key: "record_gmm-w4a16_CorrectnessTest" + depends_on: "gmm-w4a16_CorrectnessTest" + env: + CI_TARGET: "gmm-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w4a16_CorrectnessTest + + - label: "Performance tests for gmm-w4a16" + key: "gmm-w4a16_PerformanceTest" + depends_on: "record_gmm-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for gmm-w4a16" + key: "record_gmm-w4a16_PerformanceTest" + depends_on: "gmm-w4a16_PerformanceTest" + env: + CI_TARGET: "gmm-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/gmm/w4a4.yml b/.buildkite/kernel_microbenchmarks/gmm/w4a4.yml new file mode 100644 index 000000000..36273f1d5 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/gmm/w4a4.yml @@ -0,0 +1,45 @@ +# gmm-w4a4 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for gmm-w4a4" + key: "gmm-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for gmm-w4a4" + key: "record_gmm-w4a4_CorrectnessTest" + depends_on: "gmm-w4a4_CorrectnessTest" + env: + CI_TARGET: "gmm-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w4a4_CorrectnessTest + + - label: "Performance tests for gmm-w4a4" + key: "gmm-w4a4_PerformanceTest" + depends_on: "record_gmm-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for gmm-w4a4" + key: "record_gmm-w4a4_PerformanceTest" + depends_on: "gmm-w4a4_PerformanceTest" + env: + CI_TARGET: "gmm-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/gmm/w4a8.yml b/.buildkite/kernel_microbenchmarks/gmm/w4a8.yml new file mode 100644 index 000000000..f08f2d4c2 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/gmm/w4a8.yml @@ -0,0 +1,45 @@ +# gmm-w4a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for gmm-w4a8" + key: "gmm-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for gmm-w4a8" + key: "record_gmm-w4a8_CorrectnessTest" + depends_on: "gmm-w4a8_CorrectnessTest" + env: + CI_TARGET: "gmm-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w4a8_CorrectnessTest + + - label: "Performance tests for gmm-w4a8" + key: "gmm-w4a8_PerformanceTest" + depends_on: "record_gmm-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for gmm-w4a8" + key: "record_gmm-w4a8_PerformanceTest" + depends_on: "gmm-w4a8_PerformanceTest" + env: + CI_TARGET: "gmm-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/gmm/w8a16.yml b/.buildkite/kernel_microbenchmarks/gmm/w8a16.yml new file mode 100644 index 000000000..16a50ed16 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/gmm/w8a16.yml @@ -0,0 +1,45 @@ +# gmm-w8a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for gmm-w8a16" + key: "gmm-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for gmm-w8a16" + key: "record_gmm-w8a16_CorrectnessTest" + depends_on: "gmm-w8a16_CorrectnessTest" + env: + CI_TARGET: "gmm-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w8a16_CorrectnessTest + + - label: "Performance tests for gmm-w8a16" + key: "gmm-w8a16_PerformanceTest" + depends_on: "record_gmm-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for gmm-w8a16" + key: "record_gmm-w8a16_PerformanceTest" + depends_on: "gmm-w8a16_PerformanceTest" + env: + CI_TARGET: "gmm-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/gmm/w8a8.yml b/.buildkite/kernel_microbenchmarks/gmm/w8a8.yml new file mode 100644 index 000000000..51bde7e92 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/gmm/w8a8.yml @@ -0,0 +1,45 @@ +# gmm-w8a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for gmm-w8a8" + key: "gmm-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for gmm-w8a8" + key: "record_gmm-w8a8_CorrectnessTest" + depends_on: "gmm-w8a8_CorrectnessTest" + env: + CI_TARGET: "gmm-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w8a8_CorrectnessTest + + - label: "Performance tests for gmm-w8a8" + key: "gmm-w8a8_PerformanceTest" + depends_on: "record_gmm-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "gmm-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for gmm-w8a8" + key: "record_gmm-w8a8_PerformanceTest" + depends_on: "gmm-w8a8_PerformanceTest" + env: + CI_TARGET: "gmm-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh gmm-w8a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/mla/w16a16.yml b/.buildkite/kernel_microbenchmarks/mla/w16a16.yml new file mode 100644 index 000000000..a0c8df8e9 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/mla/w16a16.yml @@ -0,0 +1,45 @@ +# mla-w16a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for mla-w16a16" + key: "mla-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for mla-w16a16" + key: "record_mla-w16a16_CorrectnessTest" + depends_on: "mla-w16a16_CorrectnessTest" + env: + CI_TARGET: "mla-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w16a16_CorrectnessTest + + - label: "Performance tests for mla-w16a16" + key: "mla-w16a16_PerformanceTest" + depends_on: "record_mla-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for mla-w16a16" + key: "record_mla-w16a16_PerformanceTest" + depends_on: "mla-w16a16_PerformanceTest" + env: + CI_TARGET: "mla-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/mla/w4a16.yml b/.buildkite/kernel_microbenchmarks/mla/w4a16.yml new file mode 100644 index 000000000..a28c8456b --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/mla/w4a16.yml @@ -0,0 +1,45 @@ +# mla-w4a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for mla-w4a16" + key: "mla-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for mla-w4a16" + key: "record_mla-w4a16_CorrectnessTest" + depends_on: "mla-w4a16_CorrectnessTest" + env: + CI_TARGET: "mla-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w4a16_CorrectnessTest + + - label: "Performance tests for mla-w4a16" + key: "mla-w4a16_PerformanceTest" + depends_on: "record_mla-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for mla-w4a16" + key: "record_mla-w4a16_PerformanceTest" + depends_on: "mla-w4a16_PerformanceTest" + env: + CI_TARGET: "mla-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/mla/w4a4.yml b/.buildkite/kernel_microbenchmarks/mla/w4a4.yml new file mode 100644 index 000000000..74e226d50 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/mla/w4a4.yml @@ -0,0 +1,45 @@ +# mla-w4a4 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for mla-w4a4" + key: "mla-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for mla-w4a4" + key: "record_mla-w4a4_CorrectnessTest" + depends_on: "mla-w4a4_CorrectnessTest" + env: + CI_TARGET: "mla-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w4a4_CorrectnessTest + + - label: "Performance tests for mla-w4a4" + key: "mla-w4a4_PerformanceTest" + depends_on: "record_mla-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for mla-w4a4" + key: "record_mla-w4a4_PerformanceTest" + depends_on: "mla-w4a4_PerformanceTest" + env: + CI_TARGET: "mla-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/mla/w4a8.yml b/.buildkite/kernel_microbenchmarks/mla/w4a8.yml new file mode 100644 index 000000000..fc5219b2b --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/mla/w4a8.yml @@ -0,0 +1,45 @@ +# mla-w4a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for mla-w4a8" + key: "mla-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for mla-w4a8" + key: "record_mla-w4a8_CorrectnessTest" + depends_on: "mla-w4a8_CorrectnessTest" + env: + CI_TARGET: "mla-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w4a8_CorrectnessTest + + - label: "Performance tests for mla-w4a8" + key: "mla-w4a8_PerformanceTest" + depends_on: "record_mla-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for mla-w4a8" + key: "record_mla-w4a8_PerformanceTest" + depends_on: "mla-w4a8_PerformanceTest" + env: + CI_TARGET: "mla-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/mla/w8a16.yml b/.buildkite/kernel_microbenchmarks/mla/w8a16.yml new file mode 100644 index 000000000..d4a730268 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/mla/w8a16.yml @@ -0,0 +1,45 @@ +# mla-w8a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for mla-w8a16" + key: "mla-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for mla-w8a16" + key: "record_mla-w8a16_CorrectnessTest" + depends_on: "mla-w8a16_CorrectnessTest" + env: + CI_TARGET: "mla-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w8a16_CorrectnessTest + + - label: "Performance tests for mla-w8a16" + key: "mla-w8a16_PerformanceTest" + depends_on: "record_mla-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for mla-w8a16" + key: "record_mla-w8a16_PerformanceTest" + depends_on: "mla-w8a16_PerformanceTest" + env: + CI_TARGET: "mla-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/mla/w8a8.yml b/.buildkite/kernel_microbenchmarks/mla/w8a8.yml new file mode 100644 index 000000000..0a338a941 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/mla/w8a8.yml @@ -0,0 +1,45 @@ +# mla-w8a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for mla-w8a8" + key: "mla-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for mla-w8a8" + key: "record_mla-w8a8_CorrectnessTest" + depends_on: "mla-w8a8_CorrectnessTest" + env: + CI_TARGET: "mla-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w8a8_CorrectnessTest + + - label: "Performance tests for mla-w8a8" + key: "mla-w8a8_PerformanceTest" + depends_on: "record_mla-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "mla-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for mla-w8a8" + key: "record_mla-w8a8_PerformanceTest" + depends_on: "mla-w8a8_PerformanceTest" + env: + CI_TARGET: "mla-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh mla-w8a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w16a16.yml b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w16a16.yml new file mode 100644 index 000000000..20d05f26b --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w16a16.yml @@ -0,0 +1,45 @@ +# ragged paged attention v3 head_dim 64-w16a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for ragged paged attention v3 head_dim 64-w16a16" + key: "ragged_paged_attention_v3_head_dim_64-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w16a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for ragged paged attention v3 head_dim 64 w16a16" + key: "record_ragged_paged_attention_v3_head_dim_64-w16a16_CorrectnessTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w16a16_CorrectnessTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w16a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w16a16_CorrectnessTest + + - label: "Performance tests for ragged paged attention v3 head_dim 64-w16a16" + key: "ragged_paged_attention_v3_head_dim_64-w16a16_PerformanceTest" + depends_on: "record_ragged_paged_attention_v3_head_dim_64-w16a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w16a16_PerformanceTest" "to be added" + - label: "Record performance test result for ragged paged attention v3 head_dim 64-w16a16" + key: "record_ragged_paged_attention_v3_head_dim_64-w16a16_PerformanceTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w16a16_PerformanceTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w16a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w16a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a16.yml b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a16.yml new file mode 100644 index 000000000..802aa9755 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a16.yml @@ -0,0 +1,45 @@ +# ragged paged attention v3 head_dim 64-w4a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for ragged paged attention v3 head_dim 64-w4a16" + key: "ragged_paged_attention_v3_head_dim_64-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w4a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for ragged paged attention v3 head_dim 64 w4a16" + key: "record_ragged_paged_attention_v3_head_dim_64-w4a16_CorrectnessTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w4a16_CorrectnessTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w4a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w4a16_CorrectnessTest + + - label: "Performance tests for ragged paged attention v3 head_dim 64-w4a16" + key: "ragged_paged_attention_v3_head_dim_64-w4a16_PerformanceTest" + depends_on: "record_ragged_paged_attention_v3_head_dim_64-w4a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w4a16_PerformanceTest" "to be added" + - label: "Record performance test result for ragged paged attention v3 head_dim 64-w4a16" + key: "record_ragged_paged_attention_v3_head_dim_64-w4a16_PerformanceTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w4a16_PerformanceTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w4a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w4a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a4.yml b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a4.yml new file mode 100644 index 000000000..3529047fb --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a4.yml @@ -0,0 +1,45 @@ +# ragged paged attention v3 head_dim 64-w4a4 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for ragged paged attention v3 head_dim 64-w4a4" + key: "ragged_paged_attention_v3_head_dim_64-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w4a4_CorrectnessTest" "to be added" + - label: "Record correctness test result for ragged paged attention v3 head_dim 64 w4a4" + key: "record_ragged_paged_attention_v3_head_dim_64-w4a4_CorrectnessTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w4a4_CorrectnessTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w4a4" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w4a4_CorrectnessTest + + - label: "Performance tests for ragged paged attention v3 head_dim 64-w4a4" + key: "ragged_paged_attention_v3_head_dim_64-w4a4_PerformanceTest" + depends_on: "record_ragged_paged_attention_v3_head_dim_64-w4a4_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w4a4_PerformanceTest" "to be added" + - label: "Record performance test result for ragged paged attention v3 head_dim 64-w4a4" + key: "record_ragged_paged_attention_v3_head_dim_64-w4a4_PerformanceTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w4a4_PerformanceTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w4a4" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w4a4_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a8.yml b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a8.yml new file mode 100644 index 000000000..af76a046d --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w4a8.yml @@ -0,0 +1,45 @@ +# ragged paged attention v3 head_dim 64-w4a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for ragged paged attention v3 head_dim 64-w4a8" + key: "ragged_paged_attention_v3_head_dim_64-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w4a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for ragged paged attention v3 head_dim 64 w4a8" + key: "record_ragged_paged_attention_v3_head_dim_64-w4a8_CorrectnessTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w4a8_CorrectnessTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w4a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w4a8_CorrectnessTest + + - label: "Performance tests for ragged paged attention v3 head_dim 64-w4a8" + key: "ragged_paged_attention_v3_head_dim_64-w4a8_PerformanceTest" + depends_on: "record_ragged_paged_attention_v3_head_dim_64-w4a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w4a8_PerformanceTest" "to be added" + - label: "Record performance test result for ragged paged attention v3 head_dim 64-w4a8" + key: "record_ragged_paged_attention_v3_head_dim_64-w4a8_PerformanceTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w4a8_PerformanceTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w4a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w4a8_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w8a16.yml b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w8a16.yml new file mode 100644 index 000000000..bd3671059 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w8a16.yml @@ -0,0 +1,45 @@ +# ragged paged attention v3 head_dim 64-w8a16 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for ragged paged attention v3 head_dim 64-w8a16" + key: "ragged_paged_attention_v3_head_dim_64-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w8a16_CorrectnessTest" "to be added" + - label: "Record correctness test result for ragged paged attention v3 head_dim 64 w8a16" + key: "record_ragged_paged_attention_v3_head_dim_64-w8a16_CorrectnessTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w8a16_CorrectnessTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w8a16" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w8a16_CorrectnessTest + + - label: "Performance tests for ragged paged attention v3 head_dim 64-w8a16" + key: "ragged_paged_attention_v3_head_dim_64-w8a16_PerformanceTest" + depends_on: "record_ragged_paged_attention_v3_head_dim_64-w8a16_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w8a16_PerformanceTest" "to be added" + - label: "Record performance test result for ragged paged attention v3 head_dim 64-w8a16" + key: "record_ragged_paged_attention_v3_head_dim_64-w8a16_PerformanceTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w8a16_PerformanceTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w8a16" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w8a16_PerformanceTest diff --git a/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w8a8.yml b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w8a8.yml new file mode 100644 index 000000000..067224a64 --- /dev/null +++ b/.buildkite/kernel_microbenchmarks/ragged_paged_attention_v3_head_dim_64/w8a8.yml @@ -0,0 +1,45 @@ +# ragged paged attention v3 head_dim 64-w8a8 +# kernel support matrix microbenchmarks +steps: + - label: "Correctness tests for ragged paged attention v3 head_dim 64-w8a8" + key: "ragged_paged_attention_v3_head_dim_64-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w8a8_CorrectnessTest" "to be added" + - label: "Record correctness test result for ragged paged attention v3 head_dim 64 w8a8" + key: "record_ragged_paged_attention_v3_head_dim_64-w8a8_CorrectnessTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w8a8_CorrectnessTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w8a8" + CI_STAGE: "CorrectnessTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w8a8_CorrectnessTest + + - label: "Performance tests for ragged paged attention v3 head_dim 64-w8a8" + key: "ragged_paged_attention_v3_head_dim_64-w8a8_PerformanceTest" + depends_on: "record_ragged_paged_attention_v3_head_dim_64-w8a8_CorrectnessTest" + soft_fail: true + agents: + queue: tpu_v6e_queue + commands: + - | + buildkite-agent meta-data set "ragged_paged_attention_v3_head_dim_64-w8a8_PerformanceTest" "to be added" + - label: "Record performance test result for ragged paged attention v3 head_dim 64-w8a8" + key: "record_ragged_paged_attention_v3_head_dim_64-w8a8_PerformanceTest" + depends_on: "ragged_paged_attention_v3_head_dim_64-w8a8_PerformanceTest" + env: + CI_TARGET: "ragged paged attention v3 head_dim 64-w8a8" + CI_STAGE: "PerformanceTest" + CI_CATEGORY: "kernel support matrix microbenchmarks" + agents: + queue: cpu + commands: + - | + .buildkite/scripts/record_step_result.sh ragged_paged_attention_v3_head_dim_64-w8a8_PerformanceTest diff --git a/.buildkite/pipeline_generation/add_feature_to_ci.py b/.buildkite/pipeline_generation/add_feature_to_ci.py index 4787d2652..39a65c225 100644 --- a/.buildkite/pipeline_generation/add_feature_to_ci.py +++ b/.buildkite/pipeline_generation/add_feature_to_ci.py @@ -13,6 +13,7 @@ class FeatureCategory(str, Enum): FEATURE_SUPPORT = "feature support matrix" KERNEL_SUPPORT = "kernel support matrix" + KERNEL_SUPPORT_MICROBENCHMARKS = "kernel support matrix (microbenchmarks)" def generate_from_template(feature_name: str, feature_category: str, @@ -102,7 +103,8 @@ def main(): '--category', choices=[ FeatureCategory.FEATURE_SUPPORT.value, - FeatureCategory.KERNEL_SUPPORT.value + FeatureCategory.KERNEL_SUPPORT.value, + FeatureCategory.KERNEL_SUPPORT_MICROBENCHMARKS.value ], default='feature support matrix', help='[OPTIONAL] Category of feature. (Default: feature support matrix)' diff --git a/.buildkite/scripts/generate_support_matrices.sh b/.buildkite/scripts/generate_support_matrices.sh index 52c3aeebe..be3aed8e8 100644 --- a/.buildkite/scripts/generate_support_matrices.sh +++ b/.buildkite/scripts/generate_support_matrices.sh @@ -6,13 +6,15 @@ ANY_FAILED=false MODEL_LIST_KEY="model-list" FEATURE_LIST_KEY="feature-list" DEFAULT_FEATURES_FILE=".buildkite/features/default_features.txt" +LOCAL_TPU_VERSION="${BUILDKITE_TAG:-nightly_$(date +%Y%m%d)}" # Note: This script assumes the metadata keys contain newline-separated lists. mapfile -t model_list < <(buildkite-agent meta-data get "${MODEL_LIST_KEY}" --default "") mapfile -t metadata_feature_list < <(buildkite-agent meta-data get "${FEATURE_LIST_KEY}" --default "") MODEL_STAGES=("UnitTest" "IntegrationTest" "Benchmark") FEATURE_STAGES=("CorrectnessTest" "PerformanceTest") -FEATURE_STAGES_QUANTIZATION=("RecommendedTPUGenerations" "CorrectnessTest" "PerformanceTest") +FEATURE_STAGES_QUANTIZATION=("QuantizationMethods" "RecommendedTPUGenerations" "CorrectnessTest" "PerformanceTest") +FEATURE_STAGES_MICROBENCHMARKS=("CorrectnessTest" "PerformanceTest" "TPU Versions") declare -A TPU_GENERATIONS=( ["INT8 W8A8"]="\"v5, v6\"" @@ -22,6 +24,14 @@ declare -A TPU_GENERATIONS=( ["FP4 W4A16"]="v7" ["AWQ INT4"]="\"v5, v6\"" ) +declare -A QUANTIZATION_METHODS=( + ["INT8 W8A8"]="compressed-tensor" + ["INT4 W4A16"]="awq" + ["FP8 W8A8"]="compressed-tensor" + ["FP8 W8A16"]="compressed-tensor" + ["FP4 W4A16"]="mxfp4" + ["AWQ INT4"]="" +) declare -a model_csv_files=() declare -a feature_csv_files=() declare -a default_feature_names=() @@ -106,11 +116,16 @@ process_features() { local stages_to_use=("${FEATURE_STAGES[@]}") local header="Feature,CorrectnessTest,PerformanceTest" local is_quantization_matrix=false + local is_kernel_microbenchmarks=false if [ "$category" == "quantization support matrix" ]; then is_quantization_matrix=true stages_to_use=("${FEATURE_STAGES_QUANTIZATION[@]}") - header="Feature,Recommended TPU Generations,CorrectnessTest,PerformanceTest" + header="Quantization dtype,Quantization methods,Recommended TPU Generations,CorrectnessTest,PerformanceTest" + elif [ "$category" == "kernel support matrix microbenchmarks" ]; then + is_kernel_microbenchmarks=true + stages_to_use=("${FEATURE_STAGES_MICROBENCHMARKS[@]}") + header="kernels,CorrectnessTest,PerformanceTest,TPU Versions" fi if [ ! -f "$category_csv" ]; then @@ -123,11 +138,15 @@ process_features() { local stage_index=0 for stage in "${stages_to_use[@]}"; do local result - if [ "$is_quantization_matrix" = true ] && [ "$stage" == "RecommendedTPUGenerations" ]; then # If it's the quantization matrix, hardcode the TPU generation result="${TPU_GENERATIONS["$feature"]:-N/A}" - + elif [ "$is_quantization_matrix" = true ] && [ "$stage" == "QuantizationMethods" ]; then + # If it's the quantization matrix, hardcode the quantization methods + result="${QUANTIZATION_METHODS["$feature"]:-N/A}" + elif [ "$is_kernel_microbenchmarks" = true ] && [ "$stage" == "TPU Versions" ]; then + # If it's kernel microbenchmarks matrix, hardcode the tpu version + result="${LOCAL_TPU_VERSION}" elif [[ "$mode" == "DEFAULT" ]]; then result="✅" else @@ -136,8 +155,8 @@ process_features() { row="$row,$result" - # Check for failure (exclude the hardcoded TPU generation column) - if [ "$stage" != "RecommendedTPUGenerations" ] && [ "${result}" != "✅" ] && [ "${result}" != "N/A" ] && [ "${result}" != "to be added" ]; then + # Check for failure (exclude the hardcoded TPU generation column and Quantization Methods column) + if [ "$stage" != "TPU Versions" ] && [ "$stage" != "QuantizationMethods" ] && [ "$stage" != "RecommendedTPUGenerations" ] && [ "${result}" != "✅" ] && [ "${result}" != "N/A" ] && [ "${result}" != "to be added" ]; then ANY_FAILED=true fi @@ -147,6 +166,99 @@ process_features() { done } +process_kernel_matrix_to_pivot() { + local input_csv="kernel_support_matrix_microbenchmarks.csv" + local output_file="kernel_support_matrix-microbenchmarks.csv" + + if [ ! -f "$input_csv" ]; then + echo "Warning: Input CSV $input_csv not found. Skipping pivot." + return + fi + + # Define Headers for Display + local quant_cols_list="w16a16 w8a8 w8a16 w4a4 w4a8 w4a16" + local AWK_QUANT_COLS + AWK_QUANT_COLS=$(IFS=" "; echo "${quant_cols_list[*]}") + + # Line 1: ,w16a16,,,w8a8,,,w8a16,,,w4a4,,,w4a8,,,w4a16,, + local header_line1="," + for quant_type in $quant_cols_list; do + header_line1="${header_line1}${quant_type},,," + done + # Remove the trailing comma from the last group + header_line1="${header_line1%,}" + + # Line 2: kernels,correctness,performance,tpu versions,correctness,performance,tpu versions,... + local header_line2="kernels" + for _ in $quant_cols_list; do + header_line2="${header_line2},correctness,performance,tpu versions" + done + + # Write the two-line header structure to the output file + echo "$header_line1" > "$output_file" + echo "$header_line2" >> "$output_file" + + # Awk Script for Pivoting (Data Rows) + awk -v AWK_QUANT_COLS="$AWK_QUANT_COLS" ' + BEGIN { FS=","; OFS="," } + NR > 1 { + # Kernel parsing logic remains the same + gsub(/"/, "", $1); + + if (match($1, /-(w[0-9]+a[0-9]+)$/)) { + quant_type = substr($1, RSTART + 1, RLENGTH - 1); + base_kernel_key = substr($1, 1, RSTART - 1); + } else { + base_kernel_key = $1; + quant_type = "w16a16"; + } + + matrix[base_kernel_key][quant_type] = $2 OFS $3 OFS $4; + + if (! (base_kernel_key in kernels)) { + kernels[base_kernel_key] = 1; + kernel_list[num_kernels++] = base_kernel_key; + } + } + END { + split(AWK_QUANT_COLS, quant_cols, " "); + default_val = "N/A" OFS "N/A" OFS "N/A"; + + # Iterate through all unique base kernels found in the input (ordered by kernel_list) + for (i=0; i> "'"$output_file"'"; + } + } + ' "$input_csv" + + # Display the pivoted result + echo "--- $output_file ---" + cat "$output_file" + buildkite-agent artifact upload "$output_file" +} + if [ ${#model_list[@]} -gt 0 ]; then process_models fi @@ -173,17 +285,21 @@ done # Feature support matrices for csv_file in "${feature_csv_files[@]}"; do if [[ -f "$csv_file" ]]; then - echo "--- $csv_file ---" sorted_content=$(tail -n +2 "$csv_file" | sort -V) header=$(head -n 1 "$csv_file") echo "$header" > "$csv_file" echo "$sorted_content" >> "$csv_file" - - cat "$csv_file" - buildkite-agent artifact upload "$csv_file" + if [[ "$csv_file" != "kernel_support_matrix_microbenchmarks.csv" ]]; then + echo "--- $csv_file ---" + cat "$csv_file" + buildkite-agent artifact upload "$csv_file" + fi fi done +# Process the Kernel Matrix into the pivoted format +process_kernel_matrix_to_pivot + echo "Reports uploaded successfully." # Cleanup diff --git a/.buildkite/scripts/upload_models_and_features.sh b/.buildkite/scripts/upload_models_and_features.sh index 875177bdc..603c5d1fb 100644 --- a/.buildkite/scripts/upload_models_and_features.sh +++ b/.buildkite/scripts/upload_models_and_features.sh @@ -1,10 +1,29 @@ #!/bin/bash + BUILDKITE_DIR=".buildkite" -TARGET_FOLDERS="models features parallelism quantization" MODEL_LIST_KEY="model-list" FEATURE_LIST_KEY="feature-list" +declare -a TARGET_FOLDERS=( + "quantization" + "parallelism" + "models" + "features" +) + +# Use find to append the kernel_microbenchmarks subdirectories +KERNEL_PARENT_DIR=".buildkite/kernel_microbenchmarks" + +if [[ -d "$KERNEL_PARENT_DIR" ]]; then + while IFS= read -r dir; do + folder_path_to_add="${dir#"${BUILDKITE_DIR}"/}" + TARGET_FOLDERS+=("$folder_path_to_add") + done < <(find "$KERNEL_PARENT_DIR" -maxdepth 1 -mindepth 1 -type d) +else + echo "Warning: Kernel microbenchmarks directory '$KERNEL_PARENT_DIR' not found. Skipping dynamic folder discovery." +fi + declare -a pipeline_steps # Declare separate arrays for each list @@ -12,7 +31,7 @@ declare -a model_list declare -a feature_list -for folder_path in $TARGET_FOLDERS; do +for folder_path in "${TARGET_FOLDERS[@]}"; do folder=$BUILDKITE_DIR/$folder_path # Check if the folder exists if [[ ! -d "$folder" ]]; then @@ -37,13 +56,7 @@ for folder_path in $TARGET_FOLDERS; do "models") model_list+=("${subject_name}") ;; - "features") - feature_list+=("${subject_name}") - ;; - "parallelism") - feature_list+=("${subject_name}") - ;; - "quantization") + "features" | "parallelism" | "quantization" | "kernel_microbenchmarks"/*) feature_list+=("${subject_name}") ;; esac diff --git a/support_matrices/hardware_acceleration_and_dtype_support_matrix.csv b/support_matrices/hardware_acceleration_and_dtype_support_matrix.csv new file mode 100644 index 000000000..31952946c --- /dev/null +++ b/support_matrices/hardware_acceleration_and_dtype_support_matrix.csv @@ -0,0 +1,4 @@ +TPU Generations,Natively Supported FP4/FP8 +v5,"fp8_e4m3b11, fp8_e5m2" +v6,"fp8_e4m3b11, fp8_e5m2" +v7,"fp8_e4m3fn, fp8_e5m2, fp4" diff --git a/support_matrices/nightly/hardware_acceleration_and_dtype_support_matrix.csv b/support_matrices/nightly/hardware_acceleration_and_dtype_support_matrix.csv new file mode 100644 index 000000000..31952946c --- /dev/null +++ b/support_matrices/nightly/hardware_acceleration_and_dtype_support_matrix.csv @@ -0,0 +1,4 @@ +TPU Generations,Natively Supported FP4/FP8 +v5,"fp8_e4m3b11, fp8_e5m2" +v6,"fp8_e4m3b11, fp8_e5m2" +v7,"fp8_e4m3fn, fp8_e5m2, fp4"