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

Initialize collective profiler with model config

Browse Source
This commit is contained in:
Gahow Wang
2026-07-16 21:59:58 +08:00
parent 3121e35b0e
commit 32ab584789
3 changed files with 23 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -11,6 +11,7 @@ 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}"
NUM_TOKENS="${NUM_TOKENS:-8}"
mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw"
exec > >(tee -a "${OUTPUT_ROOT}/logs/profile.log") 2>&1
@@ -23,12 +24,13 @@ fi
export VLLM_ALLREDUCE_USE_FLASHINFER=1
export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm
echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} runtime=vLLM-0.20.0+cu129 operator=tensor_model_parallel_all_reduce backend=FlashInfer-TRTLLM tp=${TP} tokens=${NUM_TOKENS} hidden=2048 dtype=BF16 output=${OUTPUT_ROOT} expected_wall=2-6m hard_wall=720s hard_gpu_cap=${HARD_GPU_CAP}"
echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operator=tensor_model_parallel_all_reduce backend=FlashInfer-TRTLLM tp=${TP} tokens=${NUM_TOKENS} hidden=2048 dtype=BF16 output=${OUTPUT_ROOT} expected_wall=2-6m hard_wall=720s hard_gpu_cap=${HARD_GPU_CAP}"
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
git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit"
git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit"
sha256sum "${MODEL}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256"
sha256sum profile_vllm020_allreduce.py run_allreduce_profile.sh \
> "${OUTPUT_ROOT}/provenance/source.sha256"
uv pip freeze --python "${VENV_ROOT}/bin/python" \
@@ -41,6 +43,7 @@ timeout --signal=TERM --kill-after=30s 600 \
"${VENV_ROOT}/bin/torchrun" --standalone --nproc_per_node="${TP}" \
profile_vllm020_allreduce.py \
--vllm-source "${VLLM_SOURCE}" \
--model "${MODEL}" \
--output "${OUTPUT_ROOT}/raw/allreduce-tp${TP}.json" \
--num-tokens "${TOKEN_ARGS[@]}" \
--warmup-iters 3 \