Support from-onnx parameter (openvinotoolkit#2441)

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Sofya Balandina <sofya.balandina@intel.com>

Author: 3 people

tools/who_what_benchmark/whowhatbench/model_loaders.py

@@ -139,7 +139,7 @@ def load_text_model(
         from optimum.intel.openvino import OVModelForCausalLM
         try:
             model = OVModelForCausalLM.from_pretrained(
-                model_id, device=device, ov_config=ov_config
+                model_id, device=device, ov_config=ov_config, **kwargs
             )
         except Exception:
             try:

tools/who_what_benchmark/whowhatbench/wwb.py

@@ -174,6 +174,11 @@ def parse_args():
         default=42,
         help="Text-to-image specific parameter that defines the seed value.",
     )
+    parser.add_argument(
+        "--from-onnx",
+        action="store_true",
+        help="If True, the model will be loaded from ONNX format. It's converted to OpenVINO format in runtime.",
+    )
     parser.add_argument(
         "--adapters",
         type=str,
@@ -193,7 +198,6 @@ def parse_args():
         action='store_true',
         help="LLMPipeline specific parameter that defines the use of a long context prompt.",
     )
-
     parser.add_argument(
         "--empty_adapters",
         action="store_true",
@@ -586,13 +590,15 @@ def main():
     kwargs = {}
     if args.cb_config:
         kwargs["cb_config"] = read_cb_config(args.cb_config)
+    if args.from_onnx:
+        kwargs["from_onnx"] = args.from_onnx
+        kwargs["use_cache"] = False
     if args.adapters is not None:
         kwargs["adapters"] = args.adapters
         if args.alphas is not None:
             kwargs["alphas"] = args.alphas
         else:
             kwargs["alphas"] = [1.0] * len(args.adapters)
-
     kwargs["empty_adapters"] = args.empty_adapters
 
     if args.gt_data and os.path.exists(args.gt_data):