Fix GPU auto-detection for RTX 5060 Ti (sm_120) — CUDA_VISIBLE_DEVICES via run.sh -e + warm-up in Python

run.sh

@@ -247,7 +247,19 @@ if [ -n "$FILE_ARGS" ]; then
     CMD_ARGS="$CMD_ARGS --file $FILE_ARGS"
 fi
 
-docker run --rm --init $GPU_FLAG \
+# Build CUDA_VISIBLE_DEVICES env var from GPU_ARG
+CUDA_ENV_FLAG=""
+if [ -n "$GPU_ARG" ]; then
+    # Convert GPU_ARG (e.g. "0,2" or "1" or "all") to CUDA_VISIBLE_DEVICES
+    if [ "$GPU_ARG" = "all" ]; then
+        CUDA_VISIBLE="0,1,2,3"
+    else
+        CUDA_VISIBLE="$GPU_ARG"
+    fi
+    CUDA_ENV_FLAG="-e CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE}"
+fi
+
+docker run --rm --init $GPU_FLAG $CUDA_ENV_FLAG \
     --user 1000:1000 \
     -v "${INPUT_DIR}:/data/input:ro" \
     -v "${OUTPUT_DIR}:/data/output" \