import gradio as gr def get_process_config(): return { "process.numactl": gr.Checkbox( value=True, label="process.numactl", info="Runs the model with numactl", ), "process.numactl_kwargs": gr.Textbox( label="process.numactl_kwargs", value="{'cpunodebind': 0, 'membind': 0}", info="Additional python dict of kwargs to pass to numactl", ), } def get_inference_config(): return { "inference.warmup_runs": gr.Slider( step=1, value=10, minimum=0, maximum=10, label="inference.warmup_runs", info="Number of warmup runs", ), "inference.duration": gr.Slider( step=1, value=10, minimum=0, maximum=10, label="inference.duration", info="Minimum duration of the benchmark in seconds", ), "inference.iterations": gr.Slider( step=1, value=10, minimum=0, maximum=10, label="inference.iterations", info="Minimum number of iterations of the benchmark", ), "inference.latency": gr.Checkbox( value=True, label="inference.latency", info="Measures the latency of the model", ), "inference.memory": gr.Checkbox( value=True, label="inference.memory", info="Measures the peak memory consumption", ), "inference.input_shapes": gr.Textbox( label="inference.input_shapes", value="{'batch_size': 2, 'sequence_length': 16}", 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", ), } def get_pytorch_config(): return { "pytorch.torch_dtype": gr.Dropdown( value="float32", label="pytorch.torch_dtype", choices=["bfloat16", "float16", "float32", "auto"], info="The dtype to use for the model", ), "pytorch.torch_compile": gr.Checkbox( value=False, label="pytorch.torch_compile", info="Compiles the model with torch.compile", ), } def get_openvino_config(): return { "openvino.export": gr.Checkbox( value=True, label="openvino.export", info="Exports the model to ONNX", ), "openvino.use_cache": gr.Checkbox( value=True, label="openvino.use_cache", info="Uses cached ONNX model if available", ), "openvino.use_merged": gr.Checkbox( value=True, label="openvino.use_merged", info="Uses merged ONNX model if available", ), "openvino.reshape": gr.Checkbox( value=False, label="openvino.reshape", info="Reshapes the model to the input shape", ), "openvino.half": gr.Checkbox( value=False, label="openvino.half", info="Converts model to half precision", ), }