Views
No views yet
k04_fmha_prefill/ as avo_data_examples/k04_fmha_prefill_geak/, same top-level GEAK layout
as paged_attention_large._bench_common + patched kernel/test_harness.py imports (sys.path.append → task scripts/).kernel/.git omitted from the copy (cleaner worktree for AVO).kernel/.rocprofv3/ omitted.flash_attn_func in harness)1cd avo_data_examples/k04_fmha_prefill_geak
2python3 scripts/task_runner.py compile
3python3 scripts/task_runner.py correctness
4python3 scripts/task_runner.py performance1export GEAK_SUBAGENTS_ROOT=/mnt/raid0/models/avo/avo_workspace/GEAK/subagents/preprocess
2geak-avo --repo avo_data_examples/k04_fmha_prefill_geak \
3 --task "Optimize MLA FMHA prefill _attn_fwd. Metric: latency (lower is better). kernel/kernel_jit.py; optional host.py launch tuning." \
4 --test-command "python3 scripts/task_runner.py correctness && python3 scripts/task_runner.py performance" \
5 --mode full --gpu-ids 0 --no-ragORIGINAL_TASK.md.