Benchmark CPU thread settings 1 2 and 4

servers/unraid/tail2-pose-server/compose.yaml

@@ -200,6 +200,7 @@ services:
         python -c 'from tail2.models import Models; from tail2.offline_benchmark import run_offline_benchmark; m=Models(); m.load(); run_offline_benchmark(m, .35, 4)'
 
   benchmark-rtmlib-wholebody:
+    profiles: ["validation"]
     image: tail2-pose-server:0.1.0
     container_name: tail2-pose-benchmark-rtmlib-wholebody
     gpus: all
@@ -231,6 +232,37 @@ services:
       - |
         python -c 'from tail2.models import Models; from tail2.offline_benchmark import run_offline_benchmark; m=Models(); m.load(); run_offline_benchmark(m, .35, 4)'
 
+  benchmark-threads:
+    image: tail2-pose-server:0.1.0
+    container_name: tail2-pose-benchmark-threads
+    gpus: all
+    restart: "no"
+    entrypoint: ["/bin/bash", "-lc"]
+    depends_on:
+      models:
+        condition: service_completed_successfully
+    volumes:
+      - /mnt/user/appdata/tail2-pose-server/app:/app:ro
+      - /mnt/user/appdata/tail2-pose-server/models:/models:ro
+      - ./validation:/validation
+    environment:
+      PYTHONPATH: /app
+      CUDA_VISIBLE_DEVICES: "1"
+      MODEL_BACKEND: rtmlib_body
+      DETECTION_THRESHOLD: "0.35"
+      NMS_IOU_THRESHOLD: "0.60"
+      MAX_PEOPLE: "4"
+      DET_CHECKPOINT: /models/yolox-m-humanart.onnx
+      POSE_CHECKPOINT: /models/rtmpose-m-body17.onnx
+    command:
+      - |
+        set -euo pipefail
+        for threads in 1 2 4; do
+          OMP_NUM_THREADS=$$threads MKL_NUM_THREADS=$$threads OPENBLAS_NUM_THREADS=$$threads CV2_NUM_THREADS=$$threads \
+          BENCHMARK_PROFILE="rtmlib_body_threads$${threads}" BENCHMARK_OUTPUT="/validation/benchmark-threads-$${threads}.json" \
+          python -c 'from tail2.models import Models; from tail2.offline_benchmark import run_offline_benchmark; m=Models(); m.load(); run_offline_benchmark(m, .35, 4)'
+        done
+
   validate:
     profiles: ["validation"]
     image: tail2-pose-server:0.1.0