Skip to content

Commit dfbc6d3

Browse files
gedoensmaxgedoensmax
committed
refactoring logging string
1 parent 81abc62 commit dfbc6d3

6 files changed

+96
-96
lines changed

‎onnxruntime/core/providers/nv_tensorrt_rtx/nv_execution_provider.cc

+42-42
Original file line numberDiff line numberDiff line change
@@ -281,7 +281,7 @@ bool ApplyProfileShapesFromProviderOptions(std::vector<nvinfer1::IOptimizationPr
281281
std::unordered_map<std::string, std::vector<std::vector<int64_t>>>& profile_opt_shapes,
282282
ShapeRangesMap& input_explicit_shape_ranges) {
283283
if (trt_profiles.size() == 0) {
284-
LOGS_DEFAULT(WARNING) << "[Nv EP] Number of optimization profiles should be greater than 0, but it's 0.";
284+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] Number of optimization profiles should be greater than 0, but it's 0.";
285285
return false;
286286
}
287287

@@ -295,8 +295,8 @@ bool ApplyProfileShapesFromProviderOptions(std::vector<nvinfer1::IOptimizationPr
295295
input_explicit_shape_ranges[input_name] = inner_map;
296296
}
297297

298-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Begin to apply profile shapes ...";
299-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Input tensor name is '" << input_name << "', number of profiles found is " << trt_profiles.size();
298+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Begin to apply profile shapes ...";
299+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Input tensor name is '" << input_name << "', number of profiles found is " << trt_profiles.size();
300300

301301
for (size_t i = 0; i < trt_profiles.size(); i++) {
302302
nvinfer1::Dims dims = input->getDimensions();
@@ -309,7 +309,7 @@ bool ApplyProfileShapesFromProviderOptions(std::vector<nvinfer1::IOptimizationPr
309309
int shape_size = nb_dims == 0 ? 1 : static_cast<int>(profile_min_shapes[input_name][i].size());
310310
std::vector<int64_t> shapes_min(shape_size), shapes_opt(shape_size), shapes_max(shape_size);
311311

312-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] shape size of this shape tensor is " << shape_size;
312+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] shape size of this shape tensor is " << shape_size;
313313

314314
for (int j = 0; j < shape_size; j++) {
315315
auto min_value = profile_min_shapes[input_name][i][j];
@@ -318,9 +318,9 @@ bool ApplyProfileShapesFromProviderOptions(std::vector<nvinfer1::IOptimizationPr
318318
shapes_min[j] = static_cast<int64_t>(min_value);
319319
shapes_max[j] = static_cast<int64_t>(max_value);
320320
shapes_opt[j] = static_cast<int64_t>(opt_value);
321-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] shapes_min.d[" << j << "] is " << shapes_min[j];
322-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] shapes_max.d[" << j << "] is " << shapes_max[j];
323-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] shapes_opt.d[" << j << "] is " << shapes_opt[j];
321+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] shapes_min.d[" << j << "] is " << shapes_min[j];
322+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] shapes_max.d[" << j << "] is " << shapes_max[j];
323+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] shapes_opt.d[" << j << "] is " << shapes_opt[j];
324324

325325
if (input_explicit_shape_ranges[input_name].find(j) == input_explicit_shape_ranges[input_name].end()) {
326326
std::vector<std::vector<int64_t>> profile_vector(trt_profiles.size());
@@ -342,7 +342,7 @@ bool ApplyProfileShapesFromProviderOptions(std::vector<nvinfer1::IOptimizationPr
342342
dims_max.nbDims = nb_dims;
343343
dims_opt.nbDims = nb_dims;
344344

345-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] number of dimension of this execution tensor is " << nb_dims;
345+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] number of dimension of this execution tensor is " << nb_dims;
346346

347347
for (int j = 0; j < nb_dims; j++) {
348348
if (dims.d[j] == -1) {
@@ -352,9 +352,9 @@ bool ApplyProfileShapesFromProviderOptions(std::vector<nvinfer1::IOptimizationPr
352352
dims_min.d[j] = static_cast<int32_t>(min_value);
353353
dims_max.d[j] = static_cast<int32_t>(max_value);
354354
dims_opt.d[j] = static_cast<int32_t>(opt_value);
355-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] dims_min.d[" << j << "] is " << dims_min.d[j];
356-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] dims_max.d[" << j << "] is " << dims_max.d[j];
357-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] dims_opt.d[" << j << "] is " << dims_opt.d[j];
355+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] dims_min.d[" << j << "] is " << dims_min.d[j];
356+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] dims_max.d[" << j << "] is " << dims_max.d[j];
357+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] dims_opt.d[" << j << "] is " << dims_opt.d[j];
358358

359359
if (input_explicit_shape_ranges[input_name].find(j) == input_explicit_shape_ranges[input_name].end()) {
360360
std::vector<std::vector<int64_t>> profile_vector(trt_profiles.size());
@@ -933,7 +933,7 @@ NvExecutionProvider::PerThreadContext::~PerThreadContext() {
933933
bool NvExecutionProvider::PerThreadContext::CompareProfileShapes(std::string fused_node, ShapeRangesMap& shape_ranges) {
934934
if (shape_ranges.size() > 0) {
935935
if (input_shape_ranges_[fused_node] != shape_ranges) {
936-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] The shape ranges maintained by the PerThreadContext is different from the shape ranges maintained by TRT EP. \
936+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] The shape ranges maintained by the PerThreadContext is different from the shape ranges maintained by TRT EP. \
937937
This means the engine is updated and will need to update the execution context as well.";
938938
return true;
939939
}
@@ -1116,31 +1116,31 @@ NvExecutionProvider::NvExecutionProvider(const NvExecutionProviderInfo& info)
11161116
status = ParseProfileShapes(profile_min_shapes, profile_min_shapes_);
11171117
if (!status) {
11181118
profile_min_shapes_.clear();
1119-
LOGS_DEFAULT(WARNING) << "[TensorRT EP] The format of provider option 'trt_profile_min_shapes' is wrong, please follow the format of 'input1:dim1xdimd2...,input2:dim1xdim2...,...'";
1119+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] The format of provider option 'trt_profile_min_shapes' is wrong, please follow the format of 'input1:dim1xdimd2...,input2:dim1xdim2...,...'";
11201120
}
11211121
}
11221122

11231123
if (status) {
11241124
status = ParseProfileShapes(profile_max_shapes, profile_max_shapes_);
11251125
if (!status) {
11261126
profile_max_shapes_.clear();
1127-
LOGS_DEFAULT(WARNING) << "[TensorRT EP] The format of provider option 'trt_profile_max_shapes' is wrong, please follow the format of 'input1:dim1xdimd2...,input2:dim1xdim2...,...'";
1127+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] The format of provider option 'trt_profile_max_shapes' is wrong, please follow the format of 'input1:dim1xdimd2...,input2:dim1xdim2...,...'";
11281128
}
11291129
}
11301130

11311131
if (status) {
11321132
status = ParseProfileShapes(profile_opt_shapes, profile_opt_shapes_);
11331133
if (!status) {
11341134
profile_opt_shapes_.clear();
1135-
LOGS_DEFAULT(WARNING) << "[TensorRT EP] The format of provider option 'trt_profile_opt_shapes' is wrong, please follow the format of 'input1:dim1xdimd2...,input2:dim1xdim2...,...'";
1135+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] The format of provider option 'trt_profile_opt_shapes' is wrong, please follow the format of 'input1:dim1xdimd2...,input2:dim1xdim2...,...'";
11361136
}
11371137
}
11381138

11391139
if (status) {
11401140
status = ValidateProfileShapes(profile_min_shapes_, profile_max_shapes_, profile_opt_shapes_);
11411141
if (!status) {
1142-
LOGS_DEFAULT(WARNING) << "[TensorRT EP] Profile shapes validation failed. Make sure the provider options 'trt_profile_min_shapes', 'trt_profile_max_shapes' and 'trt_profile_opt_shapes' have same input name and number of profile.";
1143-
LOGS_DEFAULT(WARNING) << "[TensorRT EP] TRT EP will implicitly create optimization profiles based on input tensor for you.";
1142+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] Profile shapes validation failed. Make sure the provider options 'trt_profile_min_shapes', 'trt_profile_max_shapes' and 'trt_profile_opt_shapes' have same input name and number of profile.";
1143+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] TRT EP will implicitly create optimization profiles based on input tensor for you.";
11441144
profile_min_shapes_.clear();
11451145
profile_max_shapes_.clear();
11461146
profile_opt_shapes_.clear();
@@ -1152,11 +1152,11 @@ NvExecutionProvider::NvExecutionProvider(const NvExecutionProviderInfo& info)
11521152

11531153
// Validate setting
11541154
if (max_partition_iterations_ <= 0) {
1155-
// LOGS_DEFAULT(WARNING) << "[Nv EP] TensorRT option nv_max_partition_iterations must be a positive integer value. Set it to 1000";
1155+
// LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] TensorRT option nv_max_partition_iterations must be a positive integer value. Set it to 1000";
11561156
max_partition_iterations_ = 1000;
11571157
}
11581158
if (min_subgraph_size_ <= 0) {
1159-
// LOGS_DEFAULT(WARNING) << "[Nv EP] TensorRT option nv_min_subgraph_size must be a positive integer value. Set it to 1";
1159+
// LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] TensorRT option nv_min_subgraph_size must be a positive integer value. Set it to 1";
11601160
min_subgraph_size_ = 1;
11611161
}
11621162

@@ -1223,10 +1223,10 @@ NvExecutionProvider::NvExecutionProvider(const NvExecutionProviderInfo& info)
12231223
trt_version_ = getInferLibVersion();
12241224
CUDA_CALL_THROW(cudaRuntimeGetVersion(&cuda_version_));
12251225

1226-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] TensorRT version is " << trt_version_;
1227-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] CUDA version is " << cuda_version_;
1226+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] TensorRT version is " << trt_version_;
1227+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] CUDA version is " << cuda_version_;
12281228

1229-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Nv provider options: "
1229+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Nv provider options: "
12301230
<< "device_id: " << device_id_
12311231
<< ", nv_max_partition_iterations: " << max_partition_iterations_
12321232
<< ", nv_min_subgraph_size: " << min_subgraph_size_
@@ -1355,7 +1355,7 @@ nvinfer1::IBuilder* NvExecutionProvider::GetBuilder(TensorrtLogger& trt_logger)
13551355
void NvExecutionProvider::GetCustomOpDomainList(std::vector<OrtCustomOpDomain*>& custom_op_domain_list) const {
13561356
auto status = CreateTensorRTCustomOpDomainList(custom_op_domain_list, info_.extra_plugin_lib_paths);
13571357
if (status != Status::OK()) {
1358-
LOGS_DEFAULT(WARNING) << "[Nv EP] Failed to get TRT plugins from TRT plugin registration.";
1358+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] Failed to get TRT plugins from TRT plugin registration.";
13591359
}
13601360
}
13611361

@@ -1534,7 +1534,7 @@ std::unique_ptr<IndexedSubGraph> NvExecutionProvider::GetSubGraph(SubGraph_t gra
15341534
auto meta_def = IndexedSubGraph_MetaDef::Create();
15351535
const std::string graph_type = graph.IsSubgraph() ? "subgraph" : "graph";
15361536
meta_def->name() = "TRTKernel_" + graph_type + "_" + graph.Name() + "_" + subgraph_id;
1537-
LOGS_DEFAULT(INFO) << "[Nv EP] TensorRT subgraph MetaDef name " + meta_def->name();
1537+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] TensorRT subgraph MetaDef name " + meta_def->name();
15381538

15391539
// Assign inputs and outputs to subgraph's meta_def
15401540
for (const auto& input : inputs) {
@@ -1655,7 +1655,7 @@ SubGraphCollection_t NvExecutionProvider::GetSupportedList(SubGraphCollection_t
16551655
// Only if the newly built graph has control flow op as well as it has parent node,
16561656
// it needs to handle outer scope values before calling graph.Resolve().
16571657
if (has_control_flow_op && graph.ParentNode()) {
1658-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Handle outer scope values for the subgraph " << graph_build.Name();
1658+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Handle outer scope values for the subgraph " << graph_build.Name();
16591659
BuildSubGraphContext(graph_build);
16601660
SetGraphOuterScopeValuesAndInputs(graph_build, graph.GetGraph());
16611661
SetAllGraphInputs(graph_build);
@@ -2041,9 +2041,9 @@ NvExecutionProvider::GetCapability(const GraphViewer& graph,
20412041
}
20422042
SubGraphCollection_t consolidated_supported_nodes_vector = {{nodes_vector, true}};
20432043
if (DetectTensorRTGraphCycles(consolidated_supported_nodes_vector, graph, model_hash, false)) {
2044-
LOGS_DEFAULT(INFO) << "[Nv EP] TensorRT nodes are not consolidated because graph will have cycles after consolidation";
2044+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] TensorRT nodes are not consolidated because graph will have cycles after consolidation";
20452045
} else {
2046-
LOGS_DEFAULT(INFO) << "[Nv EP] TensorRT nodes are consolidated into one subgraph";
2046+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] TensorRT nodes are consolidated into one subgraph";
20472047
supported_nodes_vector = consolidated_supported_nodes_vector;
20482048
}
20492049
}
@@ -2108,7 +2108,7 @@ NvExecutionProvider::GetCapability(const GraphViewer& graph,
21082108
}
21092109
}
21102110
}
2111-
LOGS_DEFAULT(INFO) << "[Nv EP] Whole graph will run on Nv execution provider";
2111+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] Whole graph will run on Nv execution provider";
21122112

21132113
// The context map is only used during EP compile time, release it to save memory space.
21142114
subgraph_context_map_.clear();
@@ -2128,11 +2128,11 @@ NvExecutionProvider::GetCapability(const GraphViewer& graph,
21282128

21292129
const size_t number_of_subgraphs = supported_nodes_vector.size();
21302130
if (number_of_trt_nodes == 0) {
2131-
LOGS_DEFAULT(WARNING) << "[Nv EP] No graph will run on Nv execution provider";
2131+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] No graph will run on Nv execution provider";
21322132
} else if (number_of_trt_nodes == number_of_ort_nodes) {
2133-
LOGS_DEFAULT(INFO) << "[Nv EP] Whole graph will run on Nv execution provider";
2133+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] Whole graph will run on Nv execution provider";
21342134
} else {
2135-
LOGS_DEFAULT(INFO) << "[Nv EP] Graph is partitioned and number of subgraphs running on Nv executio provider is " << number_of_subgraphs;
2135+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] Graph is partitioned and number of subgraphs running on Nv executio provider is " << number_of_subgraphs;
21362136
}
21372137

21382138
// The context map is only used during EP compile time, release it to save memory space.
@@ -2190,20 +2190,20 @@ common::Status NvExecutionProvider::RefitEngine(std::string onnx_model_filename,
21902190
auto parser_refitter = std::unique_ptr<nvonnxparser::IParserRefitter>(
21912191
nvonnxparser::createParserRefitter(*refitter, trt_logger));
21922192
if (refit_from_file) {
2193-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Refitting from file on disk: " << onnx_model_path.string();
2193+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Refitting from file on disk: " << onnx_model_path.string();
21942194
if (!parser_refitter->refitFromFile(onnx_model_path.string().c_str())) {
21952195
return ORT_MAKE_STATUS(ONNXRUNTIME, EP_FAIL,
21962196
"Nv EP's IParserRefitter could not refit deserialized weight-stripped engine with weights contained in: " + onnx_model_path.string());
21972197
}
21982198
} else {
2199-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Refitting from byte array";
2199+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Refitting from byte array";
22002200
if (!parser_refitter->refitFromBytes(onnx_model_bytestream, onnx_model_bytestream_size)) {
22012201
return ORT_MAKE_STATUS(ONNXRUNTIME, EP_FAIL,
22022202
"Nv EP's IParserRefitter could not refit deserialized weight-stripped engine with weights contained in the provided bytestraem");
22032203
}
22042204
}
22052205
if (refitter->refitCudaEngine()) {
2206-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Successfully refitted the weight-stripped engine.";
2206+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Successfully refitted the weight-stripped engine.";
22072207
} else {
22082208
return ORT_MAKE_STATUS(ONNXRUNTIME, EP_FAIL,
22092209
"Nv EP's IRefitter could not refit deserialized weight-stripped engine with weights contained in: " + onnx_model_path.string());
@@ -2215,7 +2215,7 @@ common::Status NvExecutionProvider::RefitEngine(std::string onnx_model_filename,
22152215
nvinfer1::IHostMemory* serialized_engine = trt_engine->serialize();
22162216
std::ofstream engine_file(refitted_engine_cache, std::ios::binary | std::ios::out);
22172217
engine_file.write(reinterpret_cast<const char*>(serialized_engine->data()), serialized_engine->size());
2218-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Serialize the refitted engine to " << refitted_engine_cache;
2218+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Serialize the refitted engine to " << refitted_engine_cache;
22192219
}
22202220
return Status::OK();
22212221
}
@@ -2378,7 +2378,7 @@ Status NvExecutionProvider::CreateNodeComputeInfoFromGraph(const GraphViewer& gr
23782378
has_dynamic_shape |= tensor_is_dynamic(input);
23792379
}
23802380
if (has_dynamic_shape) {
2381-
LOGS_DEFAULT(WARNING) << "[Nv EP] No explicit optimization profile was specified. "
2381+
LOGS_DEFAULT(WARNING) << "[NvTensorRTRTX EP] No explicit optimization profile was specified. "
23822382
"We will assume a single profile with fully dynamic range. "
23832383
"This feature is experimental and may change in the future."
23842384
"If you plan to use this model as fixed shape we recommend using a free dimension override: "
@@ -2401,7 +2401,7 @@ Status NvExecutionProvider::CreateNodeComputeInfoFromGraph(const GraphViewer& gr
24012401
if (has_explicit_profile && tensor_has_profile) {
24022402
apply_profile = ApplyProfileShapesFromProviderOptions(trt_profiles, input, profile_min_shapes_, profile_max_shapes_, profile_opt_shapes_, input_explicit_shape_ranges);
24032403
} else {
2404-
LOGS_DEFAULT(INFO) << "[Nv EP] Creating implicit profile for tensor " << input_name;
2404+
LOGS_DEFAULT(INFO) << "[NvTensorRTRTX EP] Creating implicit profile for tensor " << input_name;
24052405
profile_min_shapes_[input_name] = std::vector<std::vector<int64_t>>{{}};
24062406
profile_min_shapes_[input_name][0].resize(dims.nbDims);
24072407
profile_opt_shapes_[input_name] = std::vector<std::vector<int64_t>>{{}};
@@ -2458,20 +2458,20 @@ Status NvExecutionProvider::CreateNodeComputeInfoFromGraph(const GraphViewer& gr
24582458
// enable sparse weights
24592459
if (sparsity_enable_) {
24602460
trt_config->setFlag(nvinfer1::BuilderFlag::kSPARSE_WEIGHTS);
2461-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Sparse weights are allowed";
2461+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Sparse weights are allowed";
24622462
}
24632463

24642464
// limit auxiliary streams
24652465
if (auxiliary_streams_ >= 0) {
24662466
trt_config->setMaxAuxStreams(auxiliary_streams_);
2467-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Auxiliary streams are se to " << auxiliary_streams_;
2467+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Auxiliary streams are se to " << auxiliary_streams_;
24682468
}
24692469

24702470
if (weight_stripped_engine_enable_) {
24712471
trt_config->setFlag(nvinfer1::BuilderFlag::kSTRIP_PLAN);
2472-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] STRIP_PLAN is enabled";
2472+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] STRIP_PLAN is enabled";
24732473
trt_config->setFlag(nvinfer1::BuilderFlag::kREFIT_IDENTICAL);
2474-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] REFIT_IDENTICAL is enabled";
2474+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] REFIT_IDENTICAL is enabled";
24752475
}
24762476

24772477
// Build TRT engine (if needed) and load TRT engine if:
@@ -2554,7 +2554,7 @@ Status NvExecutionProvider::CreateNodeComputeInfoFromGraph(const GraphViewer& gr
25542554
}
25552555

25562556
if (weight_stripped_engine_refit_) {
2557-
LOGS_DEFAULT(VERBOSE) << "[Nv EP] Refit engine from main ONNX file after engine build";
2557+
LOGS_DEFAULT(VERBOSE) << "[NvTensorRTRTX EP] Refit engine from main ONNX file after engine build";
25582558
char* onnx = string_buf.data();
25592559
size_t onnx_size = string_buf.size();
25602560
auto status = RefitEngine(model_path_,

0 commit comments

Comments
 (0)