Update PTQ example to fix new compile_spec requirements #1242
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Nick Comly <85702008+ncomly-nvidia@users.noreply.github.com>
github-actions
bot
added
the
documentation
|
Note: I have not run this code (don't have a system). Would be good to run it once to ensure everything is correct. |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/conversion/evaluators/aten.cpp b/tmp/changes.txt
index 4d8795f..c8c2c00 100644
--- a/workspace/core/conversion/evaluators/aten.cpp
+++ b/tmp/changes.txt
@@ -184,7 +184,7 @@ auto aten_registrations TORCHTRT_UNUSED =
int64_t start = 0;
auto startIVal = args.at(n->input(1)).IValue();
- if(!startIVal->isNone()){
+ if (!startIVal->isNone()) {
start = args.at(n->input(1)).unwrapToInt();
}
int64_t end = args.at(n->input(2)).unwrapToInt();
diff --git a/workspace/core/conversion/converters/impl/unary.cpp b/tmp/changes.txt
index a1d03a3..6b0ee2b 100644
--- a/workspace/core/conversion/converters/impl/unary.cpp
+++ b/tmp/changes.txt
@@ -10,45 +10,41 @@ namespace converters {
namespace impl {
namespace {
-
auto abs_registration TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern(
- {"aten::abs (Tensor self) -> Tensor",
- [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
- auto in = args[0].ITensor();
- bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT
- || in->getType() == nvinfer1::DataType::kHALF
- || in->getType() == nvinfer1::DataType::kINT8;
- if(unary_supported_input){
- auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
- TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
- unary_layer->setName(util::node_info(n).c_str());
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- else{
- //For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
- at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
- auto neg_one_const = tensor_to_const(ctx, neg_one);
- auto neg_layer = add_elementwise(
+ {"aten::abs (Tensor self) -> Tensor", [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
+ auto in = args[0].ITensor();
+ bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT ||
+ in->getType() == nvinfer1::DataType::kHALF || in->getType() == nvinfer1::DataType::kINT8;
+ if (unary_supported_input) {
+ auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
+ TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
+ unary_layer->setName(util::node_info(n).c_str());
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ } else {
+ // For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
+ at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
+ auto neg_one_const = tensor_to_const(ctx, neg_one);
+ auto neg_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPROD,
in,
neg_one_const,
util::node_info(n) + std::string("_Negation"));
- TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
- auto max_layer = add_elementwise(
+ TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
+ auto max_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kMAX,
in,
neg_layer->getOutput(0),
util::node_info(n) + std::string("_Max"));
- TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- }});
+ TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ }
+ }});
#define convert(unary, trt_type) \
auto unary##_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern( \
diff --git a/workspace/tests/core/conversion/converters/test_element_wise.cpp b/tmp/changes.txt
index 994fb25..540fa12 100644
--- a/workspace/tests/core/conversion/converters/test_element_wise.cpp
+++ b/tmp/changes.txt
@@ -27,8 +27,8 @@ void pointwise_test_helper(
if (!singleInput) {
torch_inputs.push_back(at::randint(1, 5, shape2, {at::kCUDA}));
}
- if(int_tensors){
- for(size_t i = 0UL; i < torch_inputs.size(); ++i){
+ if (int_tensors) {
+ for (size_t i = 0UL; i < torch_inputs.size(); ++i) {
torch_inputs[i] = torch_inputs[i].to(at::kInt);
}
}
diff --git a/workspace/tests/core/conversion/converters/test_unary.cpp b/tmp/changes.txt
index a7ab3bb..1d40c3c 100644
--- a/workspace/tests/core/conversion/converters/test_unary.cpp
+++ b/tmp/changes.txt
@@ -1,9 +1,9 @@
#include <string>
-#include "torch/torch.h"
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"
+#include "torch/torch.h"
namespace {
std::string gen_test_graph(const std::string& unary) {
@@ -22,7 +22,7 @@ TEST(Converters, ATenAbsIntConvertsCorrectly) {
auto in = at::tensor({-1, 1, -2, 2, -3, 3}, {at::kCUDA}).to(torch::kInt32);
auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
- auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
+ auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
in = at::clone(in);
params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
diff --git a/workspace/tests/core/conversion/converters/test_select.cpp b/tmp/changes.txt
index 03b6bda..67b760a 100644
--- a/workspace/tests/core/conversion/converters/test_select.cpp
+++ b/tmp/changes.txt
@@ -376,7 +376,7 @@ TEST(Converters, ATenSliceListConvertsCorrectly) {
%slice : Tensor[] = aten::slice(%list, %1, %2, %3)
%out.1 : Tensor, %out.2 : Tensor = prim::ListUnpack(%slice)
return (%out.1, %out.2))IR";
-
+
auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/conversion/evaluators/aten.cpp b/tmp/changes.txt
index 4d8795f..c8c2c00 100644
--- a/workspace/core/conversion/evaluators/aten.cpp
+++ b/tmp/changes.txt
@@ -184,7 +184,7 @@ auto aten_registrations TORCHTRT_UNUSED =
int64_t start = 0;
auto startIVal = args.at(n->input(1)).IValue();
- if(!startIVal->isNone()){
+ if (!startIVal->isNone()) {
start = args.at(n->input(1)).unwrapToInt();
}
int64_t end = args.at(n->input(2)).unwrapToInt();
diff --git a/workspace/core/conversion/converters/impl/unary.cpp b/tmp/changes.txt
index a1d03a3..6b0ee2b 100644
--- a/workspace/core/conversion/converters/impl/unary.cpp
+++ b/tmp/changes.txt
@@ -10,45 +10,41 @@ namespace converters {
namespace impl {
namespace {
-
auto abs_registration TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern(
- {"aten::abs (Tensor self) -> Tensor",
- [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
- auto in = args[0].ITensor();
- bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT
- || in->getType() == nvinfer1::DataType::kHALF
- || in->getType() == nvinfer1::DataType::kINT8;
- if(unary_supported_input){
- auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
- TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
- unary_layer->setName(util::node_info(n).c_str());
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- else{
- //For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
- at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
- auto neg_one_const = tensor_to_const(ctx, neg_one);
- auto neg_layer = add_elementwise(
+ {"aten::abs (Tensor self) -> Tensor", [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
+ auto in = args[0].ITensor();
+ bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT ||
+ in->getType() == nvinfer1::DataType::kHALF || in->getType() == nvinfer1::DataType::kINT8;
+ if (unary_supported_input) {
+ auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
+ TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
+ unary_layer->setName(util::node_info(n).c_str());
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ } else {
+ // For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
+ at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
+ auto neg_one_const = tensor_to_const(ctx, neg_one);
+ auto neg_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPROD,
in,
neg_one_const,
util::node_info(n) + std::string("_Negation"));
- TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
- auto max_layer = add_elementwise(
+ TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
+ auto max_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kMAX,
in,
neg_layer->getOutput(0),
util::node_info(n) + std::string("_Max"));
- TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- }});
+ TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ }
+ }});
#define convert(unary, trt_type) \
auto unary##_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern( \
diff --git a/workspace/tests/core/conversion/converters/test_element_wise.cpp b/tmp/changes.txt
index 994fb25..540fa12 100644
--- a/workspace/tests/core/conversion/converters/test_element_wise.cpp
+++ b/tmp/changes.txt
@@ -27,8 +27,8 @@ void pointwise_test_helper(
if (!singleInput) {
torch_inputs.push_back(at::randint(1, 5, shape2, {at::kCUDA}));
}
- if(int_tensors){
- for(size_t i = 0UL; i < torch_inputs.size(); ++i){
+ if (int_tensors) {
+ for (size_t i = 0UL; i < torch_inputs.size(); ++i) {
torch_inputs[i] = torch_inputs[i].to(at::kInt);
}
}
diff --git a/workspace/tests/core/conversion/converters/test_unary.cpp b/tmp/changes.txt
index a7ab3bb..1d40c3c 100644
--- a/workspace/tests/core/conversion/converters/test_unary.cpp
+++ b/tmp/changes.txt
@@ -1,9 +1,9 @@
#include <string>
-#include "torch/torch.h"
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"
+#include "torch/torch.h"
namespace {
std::string gen_test_graph(const std::string& unary) {
@@ -22,7 +22,7 @@ TEST(Converters, ATenAbsIntConvertsCorrectly) {
auto in = at::tensor({-1, 1, -2, 2, -3, 3}, {at::kCUDA}).to(torch::kInt32);
auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
- auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
+ auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
in = at::clone(in);
params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
diff --git a/workspace/tests/core/conversion/converters/test_select.cpp b/tmp/changes.txt
index 03b6bda..67b760a 100644
--- a/workspace/tests/core/conversion/converters/test_select.cpp
+++ b/tmp/changes.txt
@@ -376,7 +376,7 @@ TEST(Converters, ATenSliceListConvertsCorrectly) {
%slice : Tensor[] = aten::slice(%list, %1, %2, %3)
%out.1 : Tensor, %out.2 : Tensor = prim::ListUnpack(%slice)
return (%out.1, %out.2))IR";
-
+
auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());
ERROR: Some files do not conform to style guidelines
There was a problem hiding this comment.
There are some changes that do not conform to C++ style guidelines:
diff --git a/workspace/core/conversion/evaluators/aten.cpp b/tmp/changes.txt
index 4d8795f..c8c2c00 100644
--- a/workspace/core/conversion/evaluators/aten.cpp
+++ b/tmp/changes.txt
@@ -184,7 +184,7 @@ auto aten_registrations TORCHTRT_UNUSED =
int64_t start = 0;
auto startIVal = args.at(n->input(1)).IValue();
- if(!startIVal->isNone()){
+ if (!startIVal->isNone()) {
start = args.at(n->input(1)).unwrapToInt();
}
int64_t end = args.at(n->input(2)).unwrapToInt();
diff --git a/workspace/core/conversion/converters/impl/unary.cpp b/tmp/changes.txt
index a1d03a3..6b0ee2b 100644
--- a/workspace/core/conversion/converters/impl/unary.cpp
+++ b/tmp/changes.txt
@@ -10,45 +10,41 @@ namespace converters {
namespace impl {
namespace {
-
auto abs_registration TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern(
- {"aten::abs (Tensor self) -> Tensor",
- [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
- auto in = args[0].ITensor();
- bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT
- || in->getType() == nvinfer1::DataType::kHALF
- || in->getType() == nvinfer1::DataType::kINT8;
- if(unary_supported_input){
- auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
- TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
- unary_layer->setName(util::node_info(n).c_str());
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- else{
- //For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
- at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
- auto neg_one_const = tensor_to_const(ctx, neg_one);
- auto neg_layer = add_elementwise(
+ {"aten::abs (Tensor self) -> Tensor", [](ConversionCtx* ctx, const torch::jit::Node* n, args& args) -> bool {
+ auto in = args[0].ITensor();
+ bool unary_supported_input = in->getType() == nvinfer1::DataType::kFLOAT ||
+ in->getType() == nvinfer1::DataType::kHALF || in->getType() == nvinfer1::DataType::kINT8;
+ if (unary_supported_input) {
+ auto unary_layer = ctx->net->addUnary(*in, nvinfer1::UnaryOperation::kABS);
+ TORCHTRT_CHECK(unary_layer, "Unable to create abs layer from node: " << *n);
+ unary_layer->setName(util::node_info(n).c_str());
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], unary_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ } else {
+ // For types not supported by kABS, use an elementwise implementation abs(x) = max(x, -1 * x)
+ at::Tensor neg_one = torch::full({1}, -1).to(util::TRTDataTypeToScalarType(in->getType()));
+ auto neg_one_const = tensor_to_const(ctx, neg_one);
+ auto neg_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kPROD,
in,
neg_one_const,
util::node_info(n) + std::string("_Negation"));
- TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
- auto max_layer = add_elementwise(
+ TORCHTRT_CHECK(neg_layer, "Unable to create prod layer from node: " << *n);
+ auto max_layer = add_elementwise(
ctx,
nvinfer1::ElementWiseOperation::kMAX,
in,
neg_layer->getOutput(0),
util::node_info(n) + std::string("_Max"));
- TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
- auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
- LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
- return true;
- }
- }});
+ TORCHTRT_CHECK(max_layer, "Unable to create max layer from node: " << *n);
+ auto out_tensor = ctx->AssociateValueAndTensor(n->outputs()[0], max_layer->getOutput(0));
+ LOG_DEBUG("Output tensor shape: " << out_tensor->getDimensions());
+ return true;
+ }
+ }});
#define convert(unary, trt_type) \
auto unary##_registrations TORCHTRT_UNUSED = RegisterNodeConversionPatterns().pattern( \
diff --git a/workspace/tests/core/conversion/converters/test_element_wise.cpp b/tmp/changes.txt
index 994fb25..540fa12 100644
--- a/workspace/tests/core/conversion/converters/test_element_wise.cpp
+++ b/tmp/changes.txt
@@ -27,8 +27,8 @@ void pointwise_test_helper(
if (!singleInput) {
torch_inputs.push_back(at::randint(1, 5, shape2, {at::kCUDA}));
}
- if(int_tensors){
- for(size_t i = 0UL; i < torch_inputs.size(); ++i){
+ if (int_tensors) {
+ for (size_t i = 0UL; i < torch_inputs.size(); ++i) {
torch_inputs[i] = torch_inputs[i].to(at::kInt);
}
}
diff --git a/workspace/tests/core/conversion/converters/test_unary.cpp b/tmp/changes.txt
index a7ab3bb..1d40c3c 100644
--- a/workspace/tests/core/conversion/converters/test_unary.cpp
+++ b/tmp/changes.txt
@@ -1,9 +1,9 @@
#include <string>
-#include "torch/torch.h"
#include "core/compiler.h"
#include "gtest/gtest.h"
#include "tests/util/util.h"
#include "torch/csrc/jit/ir/irparser.h"
+#include "torch/torch.h"
namespace {
std::string gen_test_graph(const std::string& unary) {
@@ -22,7 +22,7 @@ TEST(Converters, ATenAbsIntConvertsCorrectly) {
auto in = at::tensor({-1, 1, -2, 2, -3, 3}, {at::kCUDA}).to(torch::kInt32);
auto params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
- auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
+ auto jit_results = torch_tensorrt::tests::util::RunGraph(g, params, {in});
in = at::clone(in);
params = torch_tensorrt::core::ir::get_static_params(g->inputs(), {});
diff --git a/workspace/tests/core/conversion/converters/test_select.cpp b/tmp/changes.txt
index 03b6bda..67b760a 100644
--- a/workspace/tests/core/conversion/converters/test_select.cpp
+++ b/tmp/changes.txt
@@ -376,7 +376,7 @@ TEST(Converters, ATenSliceListConvertsCorrectly) {
%slice : Tensor[] = aten::slice(%list, %1, %2, %3)
%out.1 : Tensor, %out.2 : Tensor = prim::ListUnpack(%slice)
return (%out.1, %out.2))IR";
-
+
auto g = std::make_shared<torch::jit::Graph>();
torch::jit::parseIR(graph, g.get());
ERROR: Some files do not conform to style guidelines
peri044
approved these changes
Aug 8, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Signed-off-by: Nick Comly 85702008+ncomly-nvidia@users.noreply.github.com
Description
Fixed
compile(model, compile_spec)to use kwargs instead. See 3557747.Another option is to just use
**compile_specbut we have been have been pushing kwargs insteadFixes # (issue)
Type of change
Please delete options that are not relevant and/or add your own.
Checklist: