gahow/aituner
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Profile FA3 KV-cache updates separately

Browse Source
This commit is contained in:
Gahow Wang
2026-07-16 21:37:01 +08:00
parent a9aed87518
commit 9bc38d8851
5 changed files with 100 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_full.sh
View File

@@ -51,7 +51,7 @@ if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then
exit 1
fi
echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=FlashAttention3 tp=${TP} cases=${#BATCH_SPECS[@]} regimes=prefill,prefix_extend,decode,true_mixed dtype=BF16 block=16 output=${OUTPUT_ROOT} expected_wall=4-10m hard_wall=1200s hard_gpu_cap=0.33_H20h"
echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operators=FlashAttention3,reshape_and_cache_flash tp=${TP} cases=${#BATCH_SPECS[@]} regimes=prefill,prefix_extend,decode,true_mixed dtype=BF16 block=16 output=${OUTPUT_ROOT} expected_wall=4-10m hard_wall=1200s hard_gpu_cap=0.33_H20h"
date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ"
nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader
@@ -78,7 +78,8 @@ timeout --signal=TERM --kill-after=30s 1080 \
--tp "${TP}" \
--batch-specs "${BATCH_SPECS[@]}" \
--warmup-iters 5 \
--repeats 10
--repeats 10 \
--profile-kv-update
test -s "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json"
sha256sum "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json" \