def main(): args = parse_args() # --------------------------- Step 1. Initialize OpenVINO Runtime Core ------------------------------------------------ log.info('Creating OpenVINO Runtime Core') core = Core() # --------------------------- Step 2. Read a model -------------------------------------------------------------------- if args.model: log.info(f'Reading the model: {args.model}') # (.xml and .bin files) or (.onnx file) model = core.read_model(args.model) # --------------------------- Step 3. Apply preprocessing ------------------------------------------------------------- if args.output_layers: output_layer_names, output_layer_ports = parse_outputs_from_args( args) model.add_outputs(list(zip(output_layer_names, output_layer_ports))) if args.layout: layouts = parse_input_layouts(args, model.inputs) ppp = PrePostProcessor(model) for i in range(len(model.inputs)): ppp.input(i).tensor().set_element_type(Type.f32) input_name = model.input(i).get_any_name() if args.layout and input_name in layouts.keys(): ppp.input(i).tensor().set_layout(Layout(layouts[input_name])) ppp.input(i).model().set_layout(Layout(layouts[input_name])) for i in range(len(model.outputs)): ppp.output(i).tensor().set_element_type(Type.f32) model = ppp.build() if args.batch_size: batch_size = args.batch_size if args.context_window_left == args.context_window_right == 0 else 1 if any([not _input.node.layout.empty for _input in model.inputs]): set_batch(model, batch_size) else: log.warning( 'Layout is not set for any input, so custom batch size is not set' ) # ---------------------------Step 4. Configure plugin --------------------------------------------------------- devices = args.device.replace('HETERO:', '').split(',') plugin_config = {} if 'GNA' in args.device: gna_device_mode = devices[0] if '_' in devices[0] else 'GNA_AUTO' devices[0] = 'GNA' plugin_config['GNA_DEVICE_MODE'] = gna_device_mode plugin_config['GNA_PRECISION'] = f'I{args.quantization_bits}' plugin_config['GNA_EXEC_TARGET'] = args.exec_target plugin_config['GNA_PWL_MAX_ERROR_PERCENT'] = str(args.pwl_me) # Set a GNA scale factor if args.import_gna_model: if args.scale_factor: log.warning( f'Custom scale factor will be used for imported GNA model: {args.import_gna_model}' ) set_scale_factors(plugin_config, parse_scale_factors(args)) else: log.info( f'Using scale factor from the imported GNA model: {args.import_gna_model}' ) else: if args.scale_factor: set_scale_factors(plugin_config, parse_scale_factors(args)) else: scale_factors = [] for file_name in re.split(', |,', args.input): _, utterances = read_utterance_file(file_name) scale_factors.append(get_scale_factor(utterances[0])) log.info( 'Using scale factor(s) calculated from first utterance') set_scale_factors(plugin_config, scale_factors) if args.export_embedded_gna_model: plugin_config[ 'GNA_FIRMWARE_MODEL_IMAGE'] = args.export_embedded_gna_model plugin_config[ 'GNA_FIRMWARE_MODEL_IMAGE_GENERATION'] = args.embedded_gna_configuration if args.performance_counter: plugin_config['PERF_COUNT'] = 'YES' device_str = f'HETERO:{",".join(devices)}' if 'HETERO' in args.device else devices[ 0] # --------------------------- Step 5. Loading model to the device ----------------------------------------------------- log.info('Loading the model to the plugin') if args.model: compiled_model = core.compile_model(model, device_str, plugin_config) else: with open(args.import_gna_model, 'rb') as f: buf = BytesIO(f.read()) compiled_model = core.import_model(buf, device_str, plugin_config) # --------------------------- Exporting GNA model using InferenceEngine AOT API --------------------------------------- if args.export_gna_model: log.info(f'Writing GNA Model to {args.export_gna_model}') user_stream = compiled_model.export_model() with open(args.export_gna_model, 'wb') as f: f.write(user_stream) return 0 if args.export_embedded_gna_model: log.info( f'Exported GNA embedded model to file {args.export_embedded_gna_model}' ) log.info( f'GNA embedded model export done for GNA generation {args.embedded_gna_configuration}' ) return 0 # --------------------------- Step 6. Set up input -------------------------------------------------------------------- if args.input_layers: input_layer_names = re.split(', |,', args.input_layers) else: input_layer_names = [ _input.any_name for _input in compiled_model.inputs ] input_file_names = re.split(', |,', args.input) if len(input_layer_names) != len(input_file_names): log.error( f'Number of model inputs ({len(compiled_model.inputs)}) is not equal ' f'to number of ark files ({len(input_file_names)})') sys.exit(-3) input_file_data = [ read_utterance_file(file_name) for file_name in input_file_names ] infer_data = [{ input_layer_names[j]: input_file_data[j].utterances[i] for j in range(len(input_layer_names)) } for i in range(len(input_file_data[0].utterances))] if args.output_layers: output_layer_names, output_layer_ports = parse_outputs_from_args(args) # If a name of output layer contains a port number then concatenate output_layer_names and output_layer_ports if ':' in compiled_model.outputs[0].any_name: output_layer_names = [ f'{output_layer_names[i]}:{output_layer_ports[i]}' for i in range(len(output_layer_names)) ] else: output_layer_names = [compiled_model.outputs[0].any_name] if args.output: output_file_names = re.split(', |,', args.output) if len(output_layer_names) != len(output_file_names): log.error( 'The number of output files is not equal to the number of model outputs.' ) sys.exit(-6) if args.reference: reference_file_names = re.split(', |,', args.reference) if len(output_layer_names) != len(reference_file_names): log.error( 'The number of reference files is not equal to the number of model outputs.' ) sys.exit(-5) reference_file_data = [ read_utterance_file(file_name) for file_name in reference_file_names ] references = [{ output_layer_names[j]: reference_file_data[j].utterances[i] for j in range(len(output_layer_names)) } for i in range(len(input_file_data[0].utterances))] # --------------------------- Step 7. Create infer request ------------------------------------------------------------ infer_request = compiled_model.create_infer_request() # --------------------------- Step 8. Do inference -------------------------------------------------------------------- log.info('Starting inference in synchronous mode') results = [] total_infer_time = 0 for i in range(len(infer_data)): start_infer_time = default_timer() # Reset states between utterance inferences to remove a memory impact for state in infer_request.query_state(): state.reset() results.append( do_inference( infer_data[i], infer_request, args.context_window_left, args.context_window_right, )) infer_time = default_timer() - start_infer_time total_infer_time += infer_time num_of_frames = infer_data[i][input_layer_names[0]].shape[0] avg_infer_time_per_frame = infer_time / num_of_frames # --------------------------- Step 9. Process output ------------------------------------------------------------------ log.info('') log.info(f'Utterance {i}:') log.info(f'Total time in Infer (HW and SW): {infer_time * 1000:.2f}ms') log.info(f'Frames in utterance: {num_of_frames}') log.info( f'Average Infer time per frame: {avg_infer_time_per_frame * 1000:.2f}ms' ) for name in output_layer_names: log.info('') log.info(f'Output blob name: {name}') log.info(f'Number scores per frame: {results[i][name].shape[1]}') if args.reference: log.info('') compare_with_reference(results[i][name], references[i][name]) if args.performance_counter: if 'GNA' in args.device: total_cycles = infer_request.profiling_info[ 0].real_time.total_seconds() stall_cycles = infer_request.profiling_info[ 1].real_time.total_seconds() active_cycles = total_cycles - stall_cycles frequency = 10**6 if args.arch == 'CORE': frequency *= GNA_CORE_FREQUENCY else: frequency *= GNA_ATOM_FREQUENCY total_inference_time = total_cycles / frequency active_time = active_cycles / frequency stall_time = stall_cycles / frequency log.info('') log.info('Performance Statistics of GNA Hardware') log.info( f' Total Inference Time: {(total_inference_time * 1000):.4f} ms' ) log.info(f' Active Time: {(active_time * 1000):.4f} ms') log.info(f' Stall Time: {(stall_time * 1000):.4f} ms') log.info('') log.info(f'Total sample time: {total_infer_time * 1000:.2f}ms') if args.output: for i, name in enumerate(output_layer_names): data = [ results[i][name] for i in range(len(input_file_data[0].utterances)) ] write_utterance_file(output_file_names[i], input_file_data[0].keys, data) log.info(f'File {output_file_names[i]} was created!') # ---------------------------------------------------------------------------------------------------------------------- log.info( 'This sample is an API example, ' 'for any performance measurements please use the dedicated benchmark_app tool\n' ) return 0
def main(): log.basicConfig(format='[ %(levelname)s ] %(message)s', level=log.INFO, stream=sys.stdout) args = parse_args() # ---------------------------Step 1. Initialize inference engine core-------------------------------------------------- log.info('Creating Inference Engine') ie = IECore() # ---------------------------Step 2. Read a model in OpenVINO Intermediate Representation--------------- if args.model: log.info(f'Reading the network: {args.model}') # .xml and .bin files net = ie.read_network(model=args.model) # ---------------------------Step 3. Configure input & output---------------------------------------------------------- log.info('Configuring input and output blobs') # Mark layers from args.output_layers as outputs if args.output_layers: net.add_outputs(get_output_layer_list(net, args, with_ports=True)) # Get names of input and output blobs input_blobs = get_input_layer_list(net, args) output_blobs = get_output_layer_list(net, args, with_ports=False) # Set input and output precision manually for blob_name in input_blobs: net.input_info[blob_name].precision = 'FP32' for blob_name in output_blobs: net.outputs[blob_name].precision = 'FP32' net.batch_size = args.batch_size # ---------------------------Step 4. Loading model to the device------------------------------------------------------- devices = args.device.replace('HETERO:', '').split(',') plugin_config = {} if 'GNA' in args.device: gna_device_mode = devices[0] if '_' in devices[0] else 'GNA_AUTO' devices[0] = 'GNA' plugin_config['GNA_DEVICE_MODE'] = gna_device_mode plugin_config['GNA_PRECISION'] = f'I{args.quantization_bits}' # Set a GNA scale factor if args.import_gna_model: if args.scale_factor: log.warning(f'Custom scale factor will be used for imported GNA model: {args.import_gna_model}') set_scale_factors(plugin_config, parse_scale_factors(args)) else: log.info(f'Using scale factor from the imported GNA model: {args.import_gna_model}') else: if args.scale_factor: set_scale_factors(plugin_config, parse_scale_factors(args)) else: scale_factors = [] for file_name in re.split(', |,', args.input): first_utterance = next(iter(read_utterance_file(file_name).values())) scale_factors.append(get_scale_factor(first_utterance)) log.info('Using scale factor(s) calculated from first utterance') set_scale_factors(plugin_config, scale_factors) if args.export_embedded_gna_model: plugin_config['GNA_FIRMWARE_MODEL_IMAGE'] = args.export_embedded_gna_model plugin_config['GNA_FIRMWARE_MODEL_IMAGE_GENERATION'] = args.embedded_gna_configuration if args.performance_counter: plugin_config['PERF_COUNT'] = 'YES' device_str = f'HETERO:{",".join(devices)}' if 'HETERO' in args.device else devices[0] log.info('Loading the model to the plugin') if args.model: exec_net = ie.load_network(net, device_str, plugin_config) else: exec_net = ie.import_network(args.import_gna_model, device_str, plugin_config) input_blobs = get_input_layer_list(exec_net, args) output_blobs = get_output_layer_list(exec_net, args, with_ports=False) if args.input: input_files = re.split(', |,', args.input) if len(input_blobs) != len(input_files): log.error(f'Number of network inputs ({len(input_blobs)}) is not equal ' f'to number of ark files ({len(input_files)})') sys.exit(-3) if args.reference: reference_files = re.split(', |,', args.reference) if len(output_blobs) != len(reference_files): log.error('The number of reference files is not equal to the number of network outputs.') sys.exit(-5) if args.output: output_files = re.split(', |,', args.output) if len(output_blobs) != len(output_files): log.error('The number of output files is not equal to the number of network outputs.') sys.exit(-6) if args.export_gna_model: log.info(f'Writing GNA Model to {args.export_gna_model}') exec_net.export(args.export_gna_model) return 0 if args.export_embedded_gna_model: log.info(f'Exported GNA embedded model to file {args.export_embedded_gna_model}') log.info(f'GNA embedded model export done for GNA generation {args.embedded_gna_configuration}') return 0 # ---------------------------Step 5. Create infer request-------------------------------------------------------------- # load_network() method of the IECore class with a specified number of requests (default 1) returns an ExecutableNetwork # instance which stores infer requests. So you already created Infer requests in the previous step. # ---------------------------Step 6. Prepare input--------------------------------------------------------------------- file_data = [read_utterance_file(file_name) for file_name in input_files] input_data = { utterance_name: { input_blobs[i]: file_data[i][utterance_name] for i in range(len(input_blobs)) } for utterance_name in file_data[0].keys() } if args.reference: references = {output_blobs[i]: read_utterance_file(reference_files[i]) for i in range(len(output_blobs))} # ---------------------------Step 7. Do inference---------------------------------------------------------------------- log.info('Starting inference in synchronous mode') results = {blob_name: {} for blob_name in output_blobs} total_infer_time = 0 for i, key in enumerate(sorted(input_data)): start_infer_time = default_timer() # Reset states between utterance inferences to remove a memory impact for request in exec_net.requests: for state in request.query_state(): state.reset() result = infer_data(input_data[key], exec_net, input_blobs, output_blobs) for blob_name in result.keys(): results[blob_name][key] = result[blob_name] infer_time = default_timer() - start_infer_time total_infer_time += infer_time num_of_frames = file_data[0][key].shape[0] avg_infer_time_per_frame = infer_time / num_of_frames # ---------------------------Step 8. Process output-------------------------------------------------------------------- log.info('') log.info(f'Utterance {i} ({key}):') log.info(f'Total time in Infer (HW and SW): {infer_time * 1000:.2f}ms') log.info(f'Frames in utterance: {num_of_frames}') log.info(f'Average Infer time per frame: {avg_infer_time_per_frame * 1000:.2f}ms') for blob_name in output_blobs: log.info('') log.info(f'Output blob name: {blob_name}') log.info(f'Number scores per frame: {results[blob_name][key].shape[1]}') if args.reference: log.info('') compare_with_reference(results[blob_name][key], references[blob_name][key]) if args.performance_counter: if 'GNA' in args.device: pc = exec_net.requests[0].get_perf_counts() total_cycles = int(pc['1.1 Total scoring time in HW']['real_time']) stall_cycles = int(pc['1.2 Stall scoring time in HW']['real_time']) active_cycles = total_cycles - stall_cycles frequency = 10**6 if args.arch == 'CORE': frequency *= GNA_CORE_FREQUENCY else: frequency *= GNA_ATOM_FREQUENCY total_inference_time = total_cycles / frequency active_time = active_cycles / frequency stall_time = stall_cycles / frequency log.info('') log.info('Performance Statistics of GNA Hardware') log.info(f' Total Inference Time: {(total_inference_time * 1000):.4f} ms') log.info(f' Active Time: {(active_time * 1000):.4f} ms') log.info(f' Stall Time: {(stall_time * 1000):.4f} ms') log.info('') log.info(f'Total sample time: {total_infer_time * 1000:.2f}ms') if args.output: for i, blob_name in enumerate(results): write_utterance_file(output_files[i], results[blob_name]) log.info(f'File {output_files[i]} was created!') # ---------------------------------------------------------------------------------------------------------------------- log.info('This sample is an API example, ' 'for any performance measurements please use the dedicated benchmark_app tool\n') return 0
def main(): log.basicConfig(format='[ %(levelname)s ] %(message)s', level=log.INFO, stream=sys.stdout) args = parse_args() # ---------------------------Step 1. Initialize inference engine core-------------------------------------------------- log.info('Creating Inference Engine') ie = IECore() # ---------------------------Step 2. Read a model in OpenVINO Intermediate Representation--------------- if args.model: log.info(f'Reading the network: {args.model}') # .xml and .bin files net = ie.read_network(model=args.model) # ---------------------------Step 3. Configure input & output---------------------------------------------------------- log.info('Configuring input and output blobs') # Get names of input and output blobs if args.input_layers: input_blobs = re.split(', |,', args.input_layers) else: input_blobs = [next(iter(net.input_info))] if args.output_layers: output_name_port = [ output.split(':') for output in re.split(', |,', args.output_layers) ] try: output_name_port = [(blob_name, int(port)) for blob_name, port in output_name_port] except ValueError: log.error('Output Parameter does not have a port.') sys.exit(-4) net.add_outputs(output_name_port) output_blobs = [blob_name for blob_name, port in output_name_port] else: output_blobs = [list(net.outputs.keys())[-1]] # Set input and output precision manually for blob_name in input_blobs: net.input_info[blob_name].precision = 'FP32' for blob_name in output_blobs: net.outputs[blob_name].precision = 'FP32' net.batch_size = args.batch_size # ---------------------------Step 4. Loading model to the device------------------------------------------------------- devices = args.device.replace('HETERO:', '').split(',') plugin_config = {} if 'GNA' in args.device: gna_device_mode = devices[0] if '_' in devices[0] else 'GNA_AUTO' devices[0] = 'GNA' plugin_config['GNA_DEVICE_MODE'] = gna_device_mode plugin_config['GNA_PRECISION'] = f'I{args.quantization_bits}' # Get a GNA scale factor if args.import_gna_model: log.info( f'Using scale factor from the imported GNA model: {args.import_gna_model}' ) else: utterances = read_utterance_file(args.input.split(',')[0]) key = sorted(utterances)[0] scale_factor = get_scale_factor(utterances[key]) log.info( f'Using scale factor of {scale_factor:.7f} calculated from first utterance.' ) plugin_config['GNA_SCALE_FACTOR'] = str(scale_factor) if args.export_embedded_gna_model: plugin_config[ 'GNA_FIRMWARE_MODEL_IMAGE'] = args.export_embedded_gna_model plugin_config[ 'GNA_FIRMWARE_MODEL_IMAGE_GENERATION'] = args.embedded_gna_configuration device_str = f'HETERO:{",".join(devices)}' if 'HETERO' in args.device else devices[ 0] log.info('Loading the model to the plugin') if args.model: exec_net = ie.load_network(net, device_str, plugin_config) else: exec_net = ie.import_network(args.import_gna_model, device_str, plugin_config) input_blobs = [next(iter(exec_net.input_info))] output_blobs = [list(exec_net.outputs.keys())[-1]] if args.input: input_files = re.split(', |,', args.input) if len(input_blobs) != len(input_files): log.error( f'Number of network inputs ({len(input_blobs)}) is not equal ' f'to number of ark files ({len(input_files)})') sys.exit(-3) if args.reference: reference_files = re.split(', |,', args.reference) if len(output_blobs) != len(reference_files): log.error( 'The number of reference files is not equal to the number of network outputs.' ) sys.exit(-5) if args.output: output_files = re.split(', |,', args.output) if len(output_blobs) != len(output_files): log.error( 'The number of output files is not equal to the number of network outputs.' ) sys.exit(-6) if args.export_gna_model: log.info(f'Writing GNA Model to {args.export_gna_model}') exec_net.export(args.export_gna_model) return 0 if args.export_embedded_gna_model: log.info( f'Exported GNA embedded model to file {args.export_embedded_gna_model}' ) log.info( f'GNA embedded model export done for GNA generation {args.embedded_gna_configuration}' ) return 0 # ---------------------------Step 5. Create infer request-------------------------------------------------------------- # load_network() method of the IECore class with a specified number of requests (default 1) returns an ExecutableNetwork # instance which stores infer requests. So you already created Infer requests in the previous step. # ---------------------------Step 6. Prepare input--------------------------------------------------------------------- file_data = [read_utterance_file(file_name) for file_name in input_files] input_data = { utterance_name: { input_blobs[i]: file_data[i][utterance_name] for i in range(len(input_blobs)) } for utterance_name in file_data[0].keys() } if args.reference: references = { output_blobs[i]: read_utterance_file(reference_files[i]) for i in range(len(output_blobs)) } # ---------------------------Step 7. Do inference---------------------------------------------------------------------- log.info('Starting inference in synchronous mode') results = {blob_name: {} for blob_name in output_blobs} infer_times = [] for key in sorted(input_data): start_infer_time = default_timer() # Reset states between utterance inferences to remove a memory impact for request in exec_net.requests: for state in request.query_state(): state.reset() result = infer_data(input_data[key], exec_net, input_blobs, output_blobs) for blob_name in result.keys(): results[blob_name][key] = result[blob_name] infer_times.append(default_timer() - start_infer_time) # ---------------------------Step 8. Process output-------------------------------------------------------------------- for blob_name in output_blobs: for i, key in enumerate(sorted(results[blob_name])): log.info(f'Utterance {i} ({key})') log.info(f'Output blob name: {blob_name}') log.info( f'Frames in utterance: {results[blob_name][key].shape[0]}') log.info( f'Total time in Infer (HW and SW): {infer_times[i] * 1000:.2f}ms' ) if args.reference: compare_with_reference(results[blob_name][key], references[blob_name][key]) log.info('') log.info(f'Total sample time: {sum(infer_times) * 1000:.2f}ms') if args.output: for i, blob_name in enumerate(results): write_utterance_file(output_files[i], results[blob_name]) log.info(f'File {output_files[i]} was created!') # ---------------------------------------------------------------------------------------------------------------------- log.info( 'This sample is an API example, ' 'for any performance measurements please use the dedicated benchmark_app tool\n' ) return 0