fix(inference): Synchronize cpu_threads configuration for HPI predictors

paddlex/inference/models/base/predictor/base_predictor.py

@@ -165,11 +165,11 @@ def __init__(
             else:
                 self._use_static_model = True
             if self._use_static_model:
-                if not use_hpip:
-                    self._pp_option = self._prepare_pp_option(pp_option, device)
-                else:
+                self._pp_option = self._prepare_pp_option(pp_option, device)
+                if use_hpip:
                     require_hpip()
                     self._hpi_config = self._prepare_hpi_config(hpi_config, device)
+                    self.sync_threads()
         else:
             self._use_hpip = False
             self._use_static_model = False
@@ -475,6 +475,15 @@ def _prepare_hpi_config(
 
         return hpi_config
 
+    def sync_threads(self):
+        if self._pp_option and self._pp_option.cpu_threads:
+            # If the user specified threads in the old system,
+            # force it into the new HPI system
+            if self._hpi_config.backend_config is None:
+                self._hpi_config.backend_config = {}
+            if "cpu_num_threads" not in self._hpi_config.backend_config:
+                self._hpi_config.backend_config["cpu_num_threads"] = self._pp_option.cpu_threads
+
     # Should this be static?
     def _get_device_info(self, device):
         if device is None:

paddlex/inference/utils/hpi.py

@@ -16,6 +16,7 @@
 import importlib.resources
 import importlib.util
 import json
+import os
 import platform
 from collections import defaultdict
 from functools import lru_cache
@@ -58,11 +59,11 @@ class HPIInfo(BaseModel):
 
 
 class OpenVINOConfig(BaseModel):
-    cpu_num_threads: int = 10
+    cpu_num_threads: int = Field(default_factory=lambda: int(os.getenv("CPU_NUM_THREADS", 10)))
 
 
 class ONNXRuntimeConfig(BaseModel):
-    cpu_num_threads: int = 10
+    cpu_num_threads: int = Field(default_factory=lambda: int(os.getenv("CPU_NUM_THREADS", 10)))
 
 
 class TensorRTConfig(BaseModel):

paddlex/inference/utils/pp_option.py

@@ -134,7 +134,7 @@ def _get_default_config(self, model_name):
             "run_mode": get_default_run_mode(model_name, device_type),
             "device_type": device_type,
             "device_id": device_id,
-            "cpu_threads": 10,
+            "cpu_threads": int(os.getenv("CPU_NUM_THREADS", 10)),
             "delete_pass": [],
             "enable_new_ir": True if model_name not in NEWIR_BLOCKLIST else False,
             "enable_cinn": False,