Skip to content

Commit f36355f

Browse files
whitneywhtsangwhitneywhtsang
authored
[CI] Rename benchmark name reported (#4250)
`attn` is confusing as now we have both flash attention and flex attention. Signed-off-by: Whitney Tsang <[email protected]>
1 parent f916e47 commit f36355f

File tree

1 file changed

+22
-23
lines changed

1 file changed

+22
-23
lines changed

‎.github/workflows/triton-benchmarks.yml‎

Lines changed: 22 additions & 23 deletions
Original file line numberDiff line numberDiff line change
@@ -244,8 +244,8 @@ jobs:
244244
python flash_attention_benchmark.py --reports $REPORTS --n_runs $N_RUNS
245245
246246
source ../../scripts/capture-hw-details.sh
247-
python build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-triton-report.csv --benchmark attn --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
248-
python build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-xetla-report.csv --benchmark attn --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
247+
python build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-triton-report.csv --benchmark flash-attn --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
248+
python build_report.py $REPORTS/attn-performance.csv $REPORTS/attn-xetla-report.csv --benchmark flash-attn --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
249249
250250
- name: Run Triton FA bwd kernel benchmark
251251
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flash_attention_bwd_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_bwd_benchmark.py') }}
@@ -256,8 +256,8 @@ jobs:
256256
mv $REPORTS/attn-performance.csv $REPORTS/attn-bwd-performance.csv
257257
258258
source ../../scripts/capture-hw-details.sh
259-
python build_report.py $REPORTS/attn-bwd-performance.csv $REPORTS/attn-bwd-triton-report.csv --benchmark attn-bwd --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
260-
python build_report.py $REPORTS/attn-bwd-performance.csv $REPORTS/attn-bwd-xetla-report.csv --benchmark attn-bwd --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
259+
python build_report.py $REPORTS/attn-bwd-performance.csv $REPORTS/attn-bwd-triton-report.csv --benchmark flash-attn-bwd --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
260+
python build_report.py $REPORTS/attn-bwd-performance.csv $REPORTS/attn-bwd-xetla-report.csv --benchmark flash-attn-bwd --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
261261
262262
- name: Run Triton FA fwd kernel benchmark - with tensor descriptors
263263
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flash_attention_tensor_desc_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_tensor_desc_benchmark.py') }}
@@ -267,22 +267,8 @@ jobs:
267267
mv $REPORTS/attn-performance.csv $REPORTS/attn-tensor-desc-performance.csv
268268
269269
source ../../scripts/capture-hw-details.sh
270-
python build_report.py $REPORTS/attn-tensor-desc-performance.csv $REPORTS/attn-tensor-desc-triton-report.csv --benchmark attn-tensor-desc --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
271-
python build_report.py $REPORTS/attn-tensor-desc-performance.csv $REPORTS/attn-tensor-desc-xetla-report.csv --benchmark attn-tensor-desc --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
272-
273-
- name: Run Prefix Sums kernel benchmark
274-
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'prefix_sums.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'prefix_sums.py') }}
275-
run: |
276-
cd benchmarks/triton_kernels_benchmark
277-
python prefix_sums.py --reports $REPORTS --n_runs $N_RUNS
278-
source ../../scripts/capture-hw-details.sh
279-
python build_report.py $REPORTS/prefix-sums.csv $REPORTS/prefix_sums-triton-report.csv --benchmark prefix_sums --compiler triton --param_cols "N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
280-
281-
- name: Run micro benchmark
282-
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'micro_benchmarks.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'micro_benchmarks') }}
283-
run: |
284-
cd benchmarks/micro_benchmarks
285-
python run_benchmarks.py --reports $REPORTS
270+
python build_report.py $REPORTS/attn-tensor-desc-performance.csv $REPORTS/attn-tensor-desc-triton-report.csv --benchmark flash-attn-tensor-desc --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
271+
python build_report.py $REPORTS/attn-tensor-desc-performance.csv $REPORTS/attn-tensor-desc-xetla-report.csv --benchmark flash-attn-tensor-desc --compiler xetla --param_cols "Z,H,N_CTX,D_HEAD,CAUSAL" --tflops_col XeTLA-TFlops --hbm_col "XeTLA-GB/s" --tag $TAG
286272
287273
- name: Run Triton FlexAttention Causal Mask fwd kernel benchmark
288274
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_benchmark_causal_mask.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_benchmark_causal_mask.py') }}
@@ -291,7 +277,7 @@ jobs:
291277
python flex_attention_benchmark_causal_mask.py --reports $REPORTS --n_runs $N_RUNS
292278
293279
source ../../scripts/capture-hw-details.sh
294-
python build_report.py $REPORTS/flexAttnCausal-performance.csv $REPORTS/flexAttnCausal-triton-report.csv --benchmark flexAttnCausal --compiler triton --param_cols "Z,H_q,H_kv,N_CTX_q,N_CTX_kv,D_HEAD_qk,D_HEAD_v" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
280+
python build_report.py $REPORTS/flexAttnCausal-performance.csv $REPORTS/flexAttnCausal-triton-report.csv --benchmark flex-attn-causal --compiler triton --param_cols "Z,H_q,H_kv,N_CTX_q,N_CTX_kv,D_HEAD_qk,D_HEAD_v" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
295281
296282
- name: Run Triton FlexAttention Custom Masks fwd kernel benchmark
297283
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_benchmark_custom_masks.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_benchmark_custom_masks.py') }}
@@ -300,9 +286,22 @@ jobs:
300286
python flex_attention_benchmark_custom_masks.py --reports $REPORTS --n_runs $N_RUNS
301287
302288
source ../../scripts/capture-hw-details.sh
303-
python build_report.py $REPORTS/flexAttnMasks-performance.csv $REPORTS/flexAttnMasks-triton-report.csv --benchmark flexAttnMasks --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,MASK" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG --mask
304-
python build_report.py $REPORTS/flexAttnMasks-performance.csv $REPORTS/flexAttnMasks-onednn-report.csv --benchmark flexAttnMasks --compiler onednn --param_cols "Z,H,N_CTX,D_HEAD,MASK" --tflops_col OneDNN-TFlops --hbm_col "OneDNN-GB/s" --tag $TAG --mask
289+
python build_report.py $REPORTS/flexAttnMasks-performance.csv $REPORTS/flexAttnMasks-triton-report.csv --benchmark flex-attn-masks --compiler triton --param_cols "Z,H,N_CTX,D_HEAD,MASK" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG --mask
290+
python build_report.py $REPORTS/flexAttnMasks-performance.csv $REPORTS/flexAttnMasks-onednn-report.csv --benchmark flex-attn-masks --compiler onednn --param_cols "Z,H,N_CTX,D_HEAD,MASK" --tflops_col OneDNN-TFlops --hbm_col "OneDNN-GB/s" --tag $TAG --mask
305291
292+
- name: Run Prefix Sums kernel benchmark
293+
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'prefix_sums.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'prefix_sums.py') }}
294+
run: |
295+
cd benchmarks/triton_kernels_benchmark
296+
python prefix_sums.py --reports $REPORTS --n_runs $N_RUNS
297+
source ../../scripts/capture-hw-details.sh
298+
python build_report.py $REPORTS/prefix-sums.csv $REPORTS/prefix_sums-triton-report.csv --benchmark prefix_sums --compiler triton --param_cols "N" --tflops_col Triton-TFlops --hbm_col "Triton-GB/s" --tag $TAG
299+
300+
- name: Run micro benchmark
301+
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'micro_benchmarks.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'micro_benchmarks') }}
302+
run: |
303+
cd benchmarks/micro_benchmarks
304+
python run_benchmarks.py --reports $REPORTS
306305
307306
- name: Upload benchmark reports
308307
if: ${{ steps.install.outcome == 'success' && !cancelled() }}

0 commit comments

Comments
 (0)