Fix conformance & e2e tests (#2540)

### Changes

- Changed `CPU_THREADS_NUM` to `INFERENCE_NUM_THREADS`;
- Renamed `accuracy_checker.py` to `ac_wrapper.py` due to import issues;

### Reason for changes

- Bugfix;

### Related tickets

- 132634

(cherry picked from commit 52b2bac)

### Tests

- post_training_quantization/309 - passed
- e2e_onnx_model_zoo/612- running

tests/onnx/test_e2e_ptq.py

@@ -184,7 +184,7 @@ def is_ov(request):
 def _read_accuracy_checker_result(root_dir: Path, key: str) -> pd.DataFrame:
     dfs = []
     for task in TASKS:
-        csv_fp = str(root_dir / task / f"accuracy_checker-{key}.csv")
+        csv_fp = str(root_dir / task / f"ac_wrapper-{key}.csv")
         dfs += [pd.read_csv(csv_fp)]
     df = pd.concat(dfs, axis=0)
     df = df[["model", "metric_value", "metric_name", "tags"]]
@@ -402,7 +402,7 @@ def test_reference_model_accuracy(
             anno_dir,
             output_dir,
             eval_size,
-            program="accuracy_checker.py",
+            program="ac_wrapper.py",
             is_quantized=False,
             is_ov_ep=False,
             is_cpu_ep=True,
@@ -445,7 +445,7 @@ def test_onnx_rt_quantized_model_accuracy(
             anno_dir,
             output_dir,
             eval_size,
-            program="accuracy_checker.py",
+            program="ac_wrapper.py",
             is_quantized=True,
             is_ov_ep=is_ov_ep,
             is_cpu_ep=is_cpu_ep,
@@ -474,7 +474,7 @@ def test_ov_quantized_model_accuracy(
             anno_dir,
             output_dir,
             eval_size,
-            program="accuracy_checker.py",
+            program="ac_wrapper.py",
             is_quantized=True,
         )
         run_command(command)

tests/post_training/pipelines/base.py

@@ -336,9 +336,9 @@ def compress(self) -> None:
         print("Quantization...")
 
         if self.backend in PT_BACKENDS:
-            cpu_threads_num = os.environ.get("CPU_THREADS_NUM")
-            if cpu_threads_num is not None:
-                torch.set_num_threads(int(cpu_threads_num))
+            inference_num_threads = os.environ.get("INFERENCE_NUM_THREADS")
+            if inference_num_threads is not None:
+                torch.set_num_threads(int(inference_num_threads))
 
         start_time = time.perf_counter()
         self.run_info.compression_memory_usage = memory_usage(self._compress, max_usage=True)

tests/post_training/pipelines/image_classification_timm.py

@@ -127,10 +127,10 @@ def _validate(self):
 
         core = ov.Core()
 
-        if os.environ.get("CPU_THREADS_NUM"):
+        if os.environ.get("INFERENCE_NUM_THREADS"):
             # Set CPU_THREADS_NUM for OpenVINO inference
-            cpu_threads_num = os.environ.get("CPU_THREADS_NUM")
-            core.set_property("CPU", properties={"CPU_THREADS_NUM": str(cpu_threads_num)})
+            inference_num_threads = os.environ.get("INFERENCE_NUM_THREADS")
+            core.set_property("CPU", properties={"INFERENCE_NUM_THREADS": str(inference_num_threads)})
 
         ov_model = core.read_model(self.path_compressed_ir)
         compiled_model = core.compile_model(ov_model)

tests/post_training/pipelines/lm_weight_compression.py

@@ -188,10 +188,10 @@ def _validate(self):
         is_stateful = self.params.get("is_stateful", False)
         core = ov.Core()
 
-        if os.environ.get("CPU_THREADS_NUM"):
+        if os.environ.get("INFERENCE_NUM_THREADS"):
             # Set CPU_THREADS_NUM for OpenVINO inference
-            cpu_threads_num = os.environ.get("CPU_THREADS_NUM")
-            core.set_property("CPU", properties={"CPU_THREADS_NUM": str(cpu_threads_num)})
+            inference_num_threads = os.environ.get("INFERENCE_NUM_THREADS")
+            core.set_property("CPU", properties={"INFERENCE_NUM_THREADS": str(inference_num_threads)})
 
         gt_data_path = TEST_ROOT / "post_training" / "data" / "wwb_ref_answers" / self.fp32_model_name / "ref_qa.csv"
         gt_data_path.parent.mkdir(parents=True, exist_ok=True)