[ci][atomesh] fix file permissions and update ATOMesh dashboard html

.github/benchmark/models_atomesh.yaml

@@ -100,6 +100,8 @@ models:
             workers: 1
             tp: 8
             extra_args: "--enable-dp-attention --enable-tbo"
+          benchmark:
+            wait_server_timeout: 4500
           run_eval: true
 
     accuracy:
@@ -145,9 +147,7 @@ models:
             decode:
               HIP_VISIBLE_DEVICES: "4,5,6,7"
           benchmark:
-            wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
+            wait_server_timeout: 4500
           run_eval: true
           eval_concurrency: [64, 256]
 
@@ -165,9 +165,7 @@ models:
             workers: 1
             tp: 4
           benchmark:
-            wait_server_timeout: 1800
-          runner:
-            time_limit: "06:00:00"
+            wait_server_timeout: 4500
           run_eval: true
           eval_concurrency: [128, 768]
 
@@ -224,8 +222,6 @@ models:
               HIP_VISIBLE_DEVICES: "4,5,6,7"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [64, 256]
 
@@ -249,8 +245,6 @@ models:
             extra_args: "--enable-dp-attention"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [256, 512, 768, 1024]
 
@@ -279,8 +273,6 @@ models:
               HIP_VISIBLE_DEVICES: "4,5,6,7"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [64, 256]
 
@@ -307,8 +299,6 @@ models:
             extra_args: "--enable-dp-attention"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [256, 512, 768, 1024]
 
@@ -370,8 +360,6 @@ models:
               HIP_VISIBLE_DEVICES: "4,5,6,7"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [64, 256]
 
@@ -396,8 +384,6 @@ models:
             extra_args: "--enable-dp-attention"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [256, 512, 768, 1024]
 
@@ -426,8 +412,6 @@ models:
               HIP_VISIBLE_DEVICES: "4,5,6,7"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [64, 256]
 
@@ -455,8 +439,6 @@ models:
             extra_args: "--enable-dp-attention"
           benchmark:
             wait_server_timeout: 1800
-          runner:
-            time_limit: "04:00:00"
           run_eval: true
           eval_concurrency: [256, 512, 768, 1024]
 

.github/scripts/atomesh_mocker_benchmark.sh → .github/scripts/atomesh/atomesh_mocker_benchmark.sh

@@ -12,7 +12,7 @@ DECODE_WORKERS="${DECODE_WORKERS:-1}"
 POLICY="${POLICY:-round_robin}"
 RESULT_DIR="${RESULT_DIR:-atomesh-mocker-results}"
 
-REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)"
+REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/../../.." && pwd)"
 MESH_DIR="${REPO_ROOT}/atom/mesh"
 MOCKER_DIR="${MESH_DIR}/mocker"
 MOCKER_TARGET_DIR="${MOCKER_DIR}/target/mocker"

.github/scripts/atomesh_mocker_benchmark_summary.py → .github/scripts/atomesh/atomesh_mocker_benchmark_summary.py

@@ -27,7 +27,7 @@ def parse_args() -> argparse.Namespace:
     )
     parser.add_argument(
         "--benchmark-script",
-        default=".github/scripts/atomesh_mocker_benchmark.sh",
+        default=".github/scripts/atomesh/atomesh_mocker_benchmark.sh",
         help="Single-cell benchmark script to invoke.",
     )
     return parser.parse_args()

.github/scripts/atomesh/pd_server_atom.sh

@@ -84,7 +84,8 @@ EVAL_CONCURRENCY="${EVAL_CONCURRENCY:-16}"
 WAIT_SERVER_TIMEOUT="${WAIT_SERVER_TIMEOUT:-2500}"
 WAIT_ROUTER_TIMEOUT="${WAIT_ROUTER_TIMEOUT:-300}"
 
-mkdir -p "${RUN_DIR}"/{logs,benchmark_results,eval_results}
+export ATOM_TORCH_PROFILER_DIR="${ATOM_TORCH_PROFILER_DIR:-${RUN_DIR}/online_quant/rank-${NODE_RANK}}"
+mkdir -p "${RUN_DIR}"/{logs,benchmark_results,eval_results} "${ATOM_TORCH_PROFILER_DIR}"
 
 role_tp="${PREFILL_TP_SIZE}"
 if [[ "${PREFILL_SINGLE_NODE_PD}" == "1" && "${NODE_RANK}" -gt 0 ]]; then

.github/workflows/atom-sglang-test.yaml

@@ -9,9 +9,9 @@ on:
       - 'docs/**'
       - 'atom/plugin/vllm/**'
       - 'atom/mesh/**'
+      - '.github/benchmark/models_atomesh.yaml'
+      - '.github/scripts/atomesh/**'
       - '.github/workflows/atomesh-*.yaml'
-      - '.github/scripts/atomesh_*.sh'
-      - '.github/dashboard/atomesh_*.html'
       - '.github/workflows/atom-vllm-*.yaml'
       - '.github/benchmark/oot_models_accuracy.json'
       - 'LICENSE'

.github/workflows/atom-test.yaml

@@ -11,9 +11,9 @@ on:
       - 'docs/**'
       - 'atom/plugin/**'
       - 'atom/mesh/**'
+      - '.github/benchmark/models_atomesh.yaml'
+      - '.github/scripts/atomesh/**'
       - '.github/workflows/atomesh-*.yaml'
-      - '.github/scripts/atomesh_*.sh'
-      - '.github/dashboard/atomesh_*.html'
       - '.github/workflows/atom-vllm-*.yaml'
       - '.github/workflows/atom-sglang-*.yaml'
       - '.github/benchmark/oot_models_accuracy.json'

.github/workflows/atom-vllm-test.yaml

@@ -9,9 +9,9 @@ on:
       - 'docs/**'
       - 'atom/plugin/sglang/**'
       - 'atom/mesh/**'
+      - '.github/benchmark/models_atomesh.yaml'
+      - '.github/scripts/atomesh/**'
       - '.github/workflows/atomesh-*.yaml'
-      - '.github/scripts/atomesh_*.sh'
-      - '.github/dashboard/atomesh_*.html'
       - '.github/workflows/atom-sglang-*.yaml'
       - '.github/benchmark/sglang_models_accuracy.json'
       - 'LICENSE'

.github/workflows/atomesh-benchmark.yaml

@@ -419,6 +419,8 @@ jobs:
           )
           PY
 
+          cp .github/dashboard/index.html /tmp/atomesh_dashboard_index.html
+
       - name: Deploy ATOMesh dashboard markdown to gh-pages
         run: |
           set -euo pipefail
@@ -429,8 +431,9 @@ jobs:
           git checkout gh-pages
           mkdir -p benchmark-dashboard
 
+          cp /tmp/atomesh_dashboard_index.html benchmark-dashboard/index.html
           cp /tmp/atomesh_summary.js benchmark-dashboard/atomesh_summary.js
 
-          git add benchmark-dashboard/atomesh_summary.js
+          git add benchmark-dashboard/index.html benchmark-dashboard/atomesh_summary.js
           git diff --cached --quiet || git commit -m "Update ATOMesh benchmark dashboard"
           git push origin gh-pages

.github/workflows/atomesh-mocker-benchmark.yaml

@@ -5,7 +5,8 @@ on:
     branches: [main]
     paths:
       - 'atom/mesh/**'
-      - '.github/scripts/atomesh_mocker_benchmark.sh'
+      - '.github/scripts/atomesh/atomesh_mocker_benchmark.sh'
+      - '.github/scripts/atomesh/atomesh_mocker_benchmark_summary.py'
       - '.github/workflows/atomesh-mocker-benchmark.yaml'
       - '.github/dashboard/atomesh_mocker_index.html'
       - 'docs/assets/atomesh_logo.png'
@@ -14,7 +15,8 @@ on:
     types: [opened, synchronize, reopened, ready_for_review]
     paths:
       - 'atom/mesh/**'
-      - '.github/scripts/atomesh_mocker_benchmark.sh'
+      - '.github/scripts/atomesh/atomesh_mocker_benchmark.sh'
+      - '.github/scripts/atomesh/atomesh_mocker_benchmark_summary.py'
       - '.github/workflows/atomesh-mocker-benchmark.yaml'
       - '.github/dashboard/atomesh_mocker_index.html'
       - 'docs/assets/atomesh_logo.png'
@@ -170,8 +172,8 @@ jobs:
           RESULT_DIR: atomesh-mocker-results
         run: |
           set -euo pipefail
-          chmod +x .github/scripts/atomesh_mocker_benchmark.sh
-          python3 .github/scripts/atomesh_mocker_benchmark_summary.py
+          chmod +x .github/scripts/atomesh/atomesh_mocker_benchmark.sh
+          python3 .github/scripts/atomesh/atomesh_mocker_benchmark_summary.py
 
       - name: Dump mocker benchmark-request log
         if: always()