Initialize vLLM config for Frontier profiling

runs/frontier-qwen30-vllm020-profile-v1/frontier_vllm020_compat.py

@@ -9,6 +9,7 @@ unchanged and adapt only that call boundary for the profiling experiment.
 from __future__ import annotations
 
 import inspect
+import os
 from typing import Any
 
 import torch
@@ -51,7 +52,19 @@ def main() -> None:
     frontier_rope._VLLM_GET_ROPE_IMPORT_ERROR = None
 
     from frontier.profiling.linear_op.main import main as frontier_main
+    from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
 
+    model_root = os.environ.get("MODEL_ROOT")
+    if not model_root:
+        raise RuntimeError("MODEL_ROOT must point to the profiled Qwen checkpoint")
+    model_config = ModelConfig(
+        model=model_root,
+        dtype="bfloat16",
+        max_model_len=8192,
+        skip_tokenizer_init=True,
+        generation_config="vllm",
+    )
+    with set_current_vllm_config(VllmConfig(model_config=model_config)):
         frontier_main()
 
 

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

@@ -16,3 +16,4 @@ OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet
 VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1"
 FRONTIER_ROOT = "/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier"
 VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build"
+MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B"

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

@@ -1,18 +1,19 @@
 version = 1
 
 [[jobs]]
-name = "qwen30-vllm020-frontier-linear-smoke-20260716-v2-rope-api"
+name = "qwen30-vllm020-frontier-linear-smoke-20260716-v3-model-context"
 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 1020 bash run_frontier_linear_smoke.sh"
-artifacts = ["artifacts/frontier-linear-smoke-v2"]
+artifacts = ["artifacts/frontier-linear-smoke-v3"]
 
 [jobs.env]
 HOME = "/tmp/wjh"
 XDG_CACHE_HOME = "/tmp/wjh/.cache"
 VLLM_CACHE_ROOT = "/tmp/wjh/.cache/vllm"
-OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/frontier-linear-smoke-v2"
+OUTPUT_ROOT = "/home/admin/cpfs/wjh/aituner/aituner-qwen30-vllm020-profile-fleet/artifacts/frontier-linear-smoke-v3"
 VENV_ROOT = "/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1"
 FRONTIER_ROOT = "/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier"
 VLLM_SOURCE = "/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build"
+MODEL_ROOT = "/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B"

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

@@ -6,6 +6,8 @@ OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}"
 VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
 FRONTIER_ROOT="${FRONTIER_ROOT:-/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier}"
 VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}"
+MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
+export MODEL_ROOT
 RUN_DIR="$(pwd -P)"
 TOKENS=(1 8 16 32 64 128 256 512 1024 2048 4096 8192)
 mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/profiles"
@@ -23,9 +25,11 @@ nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --f
 
 test "$(git -C "${FRONTIER_ROOT}" rev-parse HEAD)" = "d9cfeb6d8791fbf2f295dd9744c56a666171776e"
 test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1"
+test -s "${MODEL_ROOT}/config.json"
 git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit"
 git -C "${FRONTIER_ROOT}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit"
 git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit"
+sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256"
 sha256sum run_frontier_linear_full.sh frontier_vllm020_compat.py \
   > "${OUTPUT_ROOT}/provenance/source.sha256"
 uv pip freeze --python "${VENV_ROOT}/bin/python" \

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

@@ -6,6 +6,8 @@ OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}"
 VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
 FRONTIER_ROOT="${FRONTIER_ROOT:-/home/admin/cpfs/wjh/frontier-qwen30-vllm020-profile-v1/Frontier}"
 VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}"
+MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
+export MODEL_ROOT
 RUN_DIR="$(pwd -P)"
 mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/profiles"
 exec > >(tee -a "${OUTPUT_ROOT}/logs/smoke.log") 2>&1
@@ -22,9 +24,11 @@ nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --f
 
 test "$(git -C "${FRONTIER_ROOT}" rev-parse HEAD)" = "d9cfeb6d8791fbf2f295dd9744c56a666171776e"
 test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1"
+test -s "${MODEL_ROOT}/config.json"
 git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit"
 git -C "${FRONTIER_ROOT}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit"
 git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit"
+sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256"
 sha256sum run_frontier_linear_smoke.sh frontier_vllm020_compat.py \
   > "${OUTPUT_ROOT}/provenance/source.sha256"
 uv pip freeze --python "${VENV_ROOT}/bin/python" \