Initialize collective profiler with model config

runs/frontier-qwen30-vllm020-profile-v1/jobs_allreduce_smoke_tp2.toml

@@ -1,12 +1,12 @@
 version = 1
 
 [[jobs]]
-name = "qwen30-vllm020-flashinfer-allreduce-smoke-tp2-20260716-v2-config-context"
+name = "qwen30-vllm020-flashinfer-allreduce-smoke-tp2-20260716-v3-model-context"
 gpus = 2
 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 840 bash run_allreduce_profile.sh"
-artifacts = ["artifacts/allreduce-smoke-tp2-v2"]
+artifacts = ["artifacts/allreduce-smoke-tp2-v3"]
 
 [jobs.env]
 HOME = "/tmp/wjh"
@@ -14,6 +14,6 @@ XDG_CACHE_HOME = "/tmp/wjh/.cache"
 VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm"
 TP = "2"
 NUM_TOKENS = "8"
-OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/allreduce-smoke-tp2-v2"
+OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/allreduce-smoke-tp2-v3"
 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"

runs/frontier-qwen30-vllm020-profile-v1/profile_vllm020_allreduce.py

@@ -23,6 +23,7 @@ VLLM_COMMIT = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1"
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser()
     parser.add_argument("--vllm-source", type=Path, required=True)
+    parser.add_argument("--model", type=Path, required=True)
     parser.add_argument("--output", type=Path, required=True)
     parser.add_argument("--num-tokens", type=int, nargs="+", default=[8])
     parser.add_argument("--hidden-dim", type=int, default=2048)
@@ -69,7 +70,12 @@ def main() -> None:
         tensor_model_parallel_all_reduce,
     )
     from vllm.distributed.parallel_state import get_tp_group
-    from vllm.config import ParallelConfig, VllmConfig, set_current_vllm_config
+    from vllm.config import (
+        ModelConfig,
+        ParallelConfig,
+        VllmConfig,
+        set_current_vllm_config,
+    )
 
     rank = int(os.environ["RANK"])
     local_rank = int(os.environ["LOCAL_RANK"])
@@ -80,7 +86,15 @@ def main() -> None:
     torch.accelerator.set_device_index(device)
     torch.set_default_device(device)
     init_distributed_environment()
+    model_config = ModelConfig(
+        model=str(args.model),
+        dtype="bfloat16",
+        max_model_len=8192,
+        skip_tokenizer_init=True,
+        generation_config="vllm",
+    )
     vllm_config = VllmConfig(
+        model_config=model_config,
         parallel_config=ParallelConfig(tensor_parallel_size=world_size)
     )
     with set_current_vllm_config(vllm_config):
@@ -162,6 +176,7 @@ def main() -> None:
                 "torch_version": torch.__version__,
                 "torch_cuda": torch.version.cuda,
                 "gpu": torch.cuda.get_device_name(device),
+                "model": str(args.model),
                 "backend_env": {
                     "VLLM_ALLREDUCE_USE_FLASHINFER": "1",
                     "VLLM_FLASHINFER_ALLREDUCE_BACKEND": "trtllm",

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

@@ -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 \