Add OpenVINO GPU support (#172)

README.md

@@ -159,6 +159,7 @@ Everything else is optional or inferred at runtime, but can be configured to you
 - [x] Intel Neural Compressor backend for CPU (`backend=neural-compressor`, `backend.device=cpu`)
 - [x] TensorRT-LLM backend for CUDA (`backend=tensorrt-llm`, `backend.device=cuda`)
 - [x] OpenVINO backend for CPU (`backend=openvino`, `backend.device=cpu`)
+- [x] OpenVINO backend for GPU (`backend=openvino`, `backend.device=gpu`)
 
 ### Benchmarking 🏋️
 

optimum_benchmark/backends/config.py

@@ -61,8 +61,8 @@ def __post_init__(self):
             self.device_ids = self.device.split(":")[1]
             LOGGER.warning(f"`device` and `device_ids` are now set to `{self.device}` and `{self.device_ids}`.")
 
-        if self.device not in ["cuda", "cpu", "mps", "xla"]:
-            raise ValueError(f"`device` must be either `cuda`, `cpu`, `mps` or `xla`, but got {self.device}")
+        if self.device not in ["cuda", "cpu", "mps", "xla", "gpu"]:
+            raise ValueError(f"`device` must be either `cuda`, `cpu`, `mps`, `xla` or `gpu`, but got {self.device}")
 
         if self.device == "cuda":
             if self.device_ids is None:

optimum_benchmark/backends/openvino/backend.py

@@ -121,6 +121,7 @@ def load_ovmodel_from_pretrained(self) -> None:
             self.config.model,
             export=self.config.export,
             ov_config=self.config.openvino_config,
+            device=self.config.device,
             **self.config.hub_kwargs,
             **self.ovmodel_kwargs,
         )

optimum_benchmark/backends/openvino/config.py

@@ -37,7 +37,8 @@ class OVConfig(BackendConfig):
     def __post_init__(self):
         super().__post_init__()
 
-        if self.device != "cpu":
+        self.device = self.device.lower()
+        if self.device not in ["cpu", "gpu"]:
             raise ValueError(f"OVBackend only supports CPU devices, got {self.device}")
 
         if self.intra_op_num_threads is not None:

optimum_benchmark/trackers/latency.py

@@ -96,6 +96,8 @@ def __init__(self, device: str, backend: str):
 
         if self.backend == "pytorch" and self.device == "cuda":
             LOGGER.info("\t+ Tracking Pytorch CUDA latency")
+        elif self.backend == "openvino":
+            LOGGER.info(f"\t+ Tracking OpenVINO {self.device.upper()} latency")
         else:
             LOGGER.info("\t+ Tracking CPU latency")