add more options

app.py

@@ -1,10 +1,17 @@
 import os
 import time
-import signal # noqa
 import traceback
 
+from config_store import (
+    get_process_config,
+    get_inference_config,
+    get_openvino_config,
+    get_pytorch_config,
+)
+
 import gradio as gr
 from huggingface_hub import create_repo, whoami
+from optimum_benchmark.launchers.device_isolation_utils import *  # noqa
 from optimum_benchmark.backends.openvino.utils import TASKS_TO_OVMODEL
 from optimum_benchmark.backends.transformers_utils import TASKS_TO_MODEL_LOADERS
 from optimum_benchmark import (
@@ -17,12 +24,6 @@ from optimum_benchmark import (
 )
 from optimum_benchmark.logging_utils import setup_logging
 
-from config_store import (
-    get_process_config,
-    get_inference_config,
-    get_openvino_config,
-    get_pytorch_config,
-)
 
 DEVICE = "cpu"
 LAUNCHER = "process"
@@ -46,7 +47,7 @@ def run_benchmark(kwargs, oauth_token: gr.OAuthToken):
     token = oauth_token.token
 
     create_repo(repo_id, token=token, repo_type="dataset", exist_ok=True)
-    gr.Info(f'Created repository "{repo_id}" on the Hub.')
+    gr.Info(f'Created repository "{repo_id}" where results will be pushed.')
 
     configs = {
         "process": {},
@@ -70,7 +71,7 @@ def run_benchmark(kwargs, oauth_token: gr.OAuthToken):
 
     for key in configs.keys():
         for k, v in configs[key].items():
-            if "kwargs" in k:
+            if k in ["input_shapes", "generate_kwargs", "numactl_kwargs"]:
                 configs[key][k] = eval(v)
 
     configs["process"] = ProcessConfig(**configs.pop("process"))
@@ -169,7 +170,7 @@ def build_demo():
                 process_config = get_process_config()
 
         with gr.Row():
-            with gr.Accordion(label="Scenario Config", open=False, visible=True):
+            with gr.Accordion(label="Inference Config", open=False, visible=True):
                 inference_config = get_inference_config()
 
         with gr.Row() as backend_configs:
@@ -224,10 +225,11 @@ def build_demo():
     return demo
 
 
+demo = build_demo()
+
 if __name__ == "__main__":
     os.environ["LOG_TO_FILE"] = "0"
     os.environ["LOG_LEVEL"] = "INFO"
     setup_logging(level="INFO", prefix="MAIN-PROCESS")
 
-    demo = build_demo()
     demo.queue(max_size=10).launch()

config_store.py

@@ -52,6 +52,16 @@ def get_inference_config():
             label="inference.memory",
             info="Measures the peak memory consumption",
         ),
+        "inference.input_shapes": gr.Textbox(
+            label="inference.input_shapes",
+            value="{'batch_size': 1, 'sequence_length': 128}",
+            info="Input shapes to use for the benchmark",
+        ),
+        "inference.generate_kwargs": gr.Textbox(
+            label="inference.generate_kwargs",
+            value="{'max_new_tokens': 32, 'min_new_tokens': 32}",
+            info="Additional python dict of kwargs to pass to the generate function",
+        ),
     }