Add full per-TP FlashAttention profile sweep

runs/frontier-qwen30-vllm020-profile-v1/jobs_flashattn_full.toml

@@ -0,0 +1,55 @@
+version = 1
+
+[[jobs]]
+name = "qwen30-vllm020-flashattn-full-tp1-20260716-v1"
+gpus = 1
+gpu_model = "H20"
+hosts = ["dash0"]
+command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1320 bash run_flashattn_full.sh"
+artifacts = ["artifacts/flashattn-full-tp1"]
+
+[jobs.env]
+HOME = "/tmp/wjh"
+XDG_CACHE_HOME = "/tmp/wjh/.cache"
+VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm"
+TP = "1"
+OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-full-tp1"
+VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1"
+VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build"
+MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B"
+
+[[jobs]]
+name = "qwen30-vllm020-flashattn-full-tp2-20260716-v1"
+gpus = 1
+gpu_model = "H20"
+hosts = ["dash0"]
+command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1320 bash run_flashattn_full.sh"
+artifacts = ["artifacts/flashattn-full-tp2"]
+
+[jobs.env]
+HOME = "/tmp/wjh"
+XDG_CACHE_HOME = "/tmp/wjh/.cache"
+VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm"
+TP = "2"
+OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-full-tp2"
+VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1"
+VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build"
+MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B"
+
+[[jobs]]
+name = "qwen30-vllm020-flashattn-full-tp4-20260716-v1"
+gpus = 1
+gpu_model = "H20"
+hosts = ["dash0"]
+command = "cd /home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-v1/runs/frontier-qwen30-vllm020-profile-v1 && timeout --signal=TERM --kill-after=30s 1320 bash run_flashattn_full.sh"
+artifacts = ["artifacts/flashattn-full-tp4"]
+
+[jobs.env]
+HOME = "/tmp/wjh"
+XDG_CACHE_HOME = "/tmp/wjh/.cache"
+VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm"
+TP = "4"
+OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/flashattn-full-tp4"
+VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1"
+VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build"
+MODEL = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B"

runs/frontier-qwen30-vllm020-profile-v1/run_flashattn_full.sh

@@ -0,0 +1,88 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+TP="${TP:?TP must be set to 1, 2, or 4}"
+case "${TP}" in
+  1|2|4) ;;
+  *) echo "ERROR: invalid TP=${TP}" >&2; exit 1 ;;
+esac
+
+OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}"
+VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
+VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}"
+MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
+LOG_DIR="${OUTPUT_ROOT}/logs"
+PROVENANCE_DIR="${OUTPUT_ROOT}/provenance"
+
+PREFILL_SPECS=(q64 q128 q256 q512 q1k q2k q4k q8k)
+PREFIX_EXTEND_SPECS=(q64s512 q128s1k q256s2k q512s4k q1ks8k)
+DECODE_SPECS=(
+  q1s128 8q1s128 16q1s128 32q1s128 64q1s128
+  8q1s1k 16q1s1k 32q1s1k 64q1s1k
+  8q1s2k 16q1s2k 32q1s2k 64q1s2k
+  8q1s4k 16q1s4k 32q1s4k 64q1s4k
+  8q1s8k 16q1s8k 32q1s8k 64q1s8k
+)
+MIXED_SPECS=(
+  q64_8q1s512
+  q128_8q1s1k q128_16q1s1k
+  q256_16q1s2k q256_32q1s2k
+  q512_32q1s4k q512_64q1s4k
+  q1k_64q1s8k q2k_32q1s4k q4k_16q1s4k
+)
+BATCH_SPECS=(
+  "${PREFILL_SPECS[@]}"
+  "${PREFIX_EXTEND_SPECS[@]}"
+  "${DECODE_SPECS[@]}"
+  "${MIXED_SPECS[@]}"
+)
+
+mkdir -p "${LOG_DIR}" "${PROVENANCE_DIR}" "${OUTPUT_ROOT}/raw"
+exec > >(tee -a "${LOG_DIR}/full.log") 2>&1
+
+if [[ -z "${CUDA_VISIBLE_DEVICES:-}" ]]; then
+  echo "ERROR: CUDA_VISIBLE_DEVICES must contain the fleet-allocated GPU" >&2
+  exit 1
+fi
+IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES}"
+if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then
+  echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2
+  exit 1
+fi
+
+echo "PROFILE_LAUNCH_ECHO host=$(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=${OUTPUT_ROOT} expected_wall=4-10m hard_wall=1200s hard_gpu_cap=0.33_H20h"
+date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ"
+nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader
+
+test -x "${VENV_ROOT}/bin/python"
+test -f "${VLLM_SOURCE}/benchmarks/attention_benchmarks/runner.py"
+test -f "${MODEL}/config.json"
+test -f profile_vllm020_flashattn.py
+
+git rev-parse HEAD > "${PROVENANCE_DIR}/aituner.commit"
+git -C "${VLLM_SOURCE}" rev-parse HEAD > "${PROVENANCE_DIR}/vllm-source.commit"
+sha256sum profile_vllm020_flashattn.py run_flashattn_full.sh \
+  > "${PROVENANCE_DIR}/source.sha256"
+uv pip freeze --python "${VENV_ROOT}/bin/python" \
+  > "${PROVENANCE_DIR}/pip-freeze.txt"
+nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \
+  --format=csv,noheader > "${PROVENANCE_DIR}/gpus.csv"
+printf '%s\n' "${BATCH_SPECS[@]}" > "${PROVENANCE_DIR}/batch-specs.txt"
+
+timeout --signal=TERM --kill-after=30s 1080 \
+  "${VENV_ROOT}/bin/python" profile_vllm020_flashattn.py \
+  --vllm-source "${VLLM_SOURCE}" \
+  --model "${MODEL}" \
+  --output "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json" \
+  --tp "${TP}" \
+  --batch-specs "${BATCH_SPECS[@]}" \
+  --warmup-iters 5 \
+  --repeats 10
+
+test -s "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json"
+sha256sum "${OUTPUT_ROOT}/raw/flashattn-tp${TP}.json" \
+  "${PROVENANCE_DIR}"/* > "${OUTPUT_ROOT}/artifacts.sha256"
+nvidia-smi --query-gpu=index,name,memory.used,utilization.gpu --format=csv,noheader
+date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ"
+echo "FLASHATTN_FULL_COMPLETE tp=${TP} cases=${#BATCH_SPECS[@]}"