New API to register arbitrary TRT engines as TorchScript modules #431
Conversation
This file contains hidden or 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
github-actions
bot
added
component: api [Python]
![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/var/Var.cpp b/tmp/changes.txt
index 71b77f5..8356934 100644
--- a/workspace/core/conversion/var/Var.cpp
+++ b/tmp/changes.txt
@@ -89,7 +89,7 @@ nvinfer1::ITensor* Var::ITensorOrFreeze(ConversionCtx* ctx) {
if (isIValue()) {
LOG_DEBUG(ctx->logger, "Found IValue containing object of type " << *(ptr_.ivalue->type()));
}
-
+
TRTORCH_CHECK(
isITensor() || (isIValue() && (ptr_.ivalue->isTensor() || ptr_.ivalue->isCustomClass())),
"Requested either IValue containing a Tensor, or ITensor, however Var type is " << type_name());
@@ -100,8 +100,10 @@ nvinfer1::ITensor* Var::ITensorOrFreeze(ConversionCtx* ctx) {
if (ptr_.ivalue->isTensor()) {
auto weights = converters::Weights();
auto tensor = ptr_.ivalue->toTensor();
- if ((tensor.scalar_type() == at::kLong || tensor.scalar_type() == at::kDouble) && !ctx->settings.truncate_long_and_double) {
- TRTORCH_THROW_ERROR("Unable to freeze tensor of type Int64/Float64 into constant layer, try to compile model with truncate_long_and_double enabled");
+ if ((tensor.scalar_type() == at::kLong || tensor.scalar_type() == at::kDouble) &&
+ !ctx->settings.truncate_long_and_double) {
+ TRTORCH_THROW_ERROR(
+ "Unable to freeze tensor of type Int64/Float64 into constant layer, try to compile model with truncate_long_and_double enabled");
} else if (tensor.scalar_type() == at::kLong && ctx->settings.truncate_long_and_double) {
weights = converters::Weights(ctx, tensor.toType(at::kInt));
LOG_WARNING("Truncating weight (constant in the graph) from Int64 to Int32");
@@ -111,7 +113,7 @@ nvinfer1::ITensor* Var::ITensorOrFreeze(ConversionCtx* ctx) {
} else {
weights = converters::Weights(ctx, tensor);
}
-
+
auto const_layer = ctx->net->addConstant(weights.shape, weights.data);
TRTORCH_CHECK(const_layer, "Unable to freeze tensor into constant layer");
out = const_layer->getOutput(0);
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/var/Var.cpp b/tmp/changes.txt
index 71b77f5..8356934 100644
--- a/workspace/core/conversion/var/Var.cpp
+++ b/tmp/changes.txt
@@ -89,7 +89,7 @@ nvinfer1::ITensor* Var::ITensorOrFreeze(ConversionCtx* ctx) {
if (isIValue()) {
LOG_DEBUG(ctx->logger, "Found IValue containing object of type " << *(ptr_.ivalue->type()));
}
-
+
TRTORCH_CHECK(
isITensor() || (isIValue() && (ptr_.ivalue->isTensor() || ptr_.ivalue->isCustomClass())),
"Requested either IValue containing a Tensor, or ITensor, however Var type is " << type_name());
@@ -100,8 +100,10 @@ nvinfer1::ITensor* Var::ITensorOrFreeze(ConversionCtx* ctx) {
if (ptr_.ivalue->isTensor()) {
auto weights = converters::Weights();
auto tensor = ptr_.ivalue->toTensor();
- if ((tensor.scalar_type() == at::kLong || tensor.scalar_type() == at::kDouble) && !ctx->settings.truncate_long_and_double) {
- TRTORCH_THROW_ERROR("Unable to freeze tensor of type Int64/Float64 into constant layer, try to compile model with truncate_long_and_double enabled");
+ if ((tensor.scalar_type() == at::kLong || tensor.scalar_type() == at::kDouble) &&
+ !ctx->settings.truncate_long_and_double) {
+ TRTORCH_THROW_ERROR(
+ "Unable to freeze tensor of type Int64/Float64 into constant layer, try to compile model with truncate_long_and_double enabled");
} else if (tensor.scalar_type() == at::kLong && ctx->settings.truncate_long_and_double) {
weights = converters::Weights(ctx, tensor.toType(at::kInt));
LOG_WARNING("Truncating weight (constant in the graph) from Int64 to Int32");
@@ -111,7 +113,7 @@ nvinfer1::ITensor* Var::ITensorOrFreeze(ConversionCtx* ctx) {
} else {
weights = converters::Weights(ctx, tensor);
}
-
+
auto const_layer = ctx->net->addConstant(weights.shape, weights.data);
TRTORCH_CHECK(const_layer, "Unable to freeze tensor into constant layer");
out = const_layer->getOutput(0);
ERROR: Some files do not conform to style guidelines
Modules Signed-off-by: Naren Dasan <[email protected]> Signed-off-by: Naren Dasan <[email protected]>
Also adds tests to confirm TRT Python API intercompatiability Signed-off-by: Naren Dasan <[email protected]> Signed-off-by: Naren Dasan <[email protected]>
narendasan
force-pushed
the
arbitrary_trt_engines
branch
from
ff8b6bc to
88d07a9
Compare
There was a problem hiding this comment.
There are some changes that do not conform to Python style guidelines:
Reformatting /workspace/cpp/ptq/training/vgg16/export_ckpt.py
Reformatting /workspace/cpp/ptq/training/vgg16/vgg16.py
Reformatting /workspace/cpp/ptq/training/vgg16/main.py
Reformatting /workspace/py/trtorch/ptq.py
Reformatting /workspace/py/trtorch/_compiler.py
Reformatting /workspace/py/trtorch/_compile_spec.py
Reformatting /workspace/py/trtorch/logging.py
Reformatting /workspace/py/trtorch/_types.py
Reformatting /workspace/py/trtorch/__init__.py
Reformatting /workspace/py/setup.py
--- /workspace/tests/py/test_api.py (original)
+++ /workspace/tests/py/test_api.py (reformatted)
@@ -45,7 +45,9 @@
same = (trt_mod(self.input) - self.scripted_model(self.input)).abs().max()
self.assertTrue(same < 2e-3)
+
class TestPTtoTRTtoPT(ModelTestCase):
+
def setUp(self):
self.input = torch.randn((1, 3, 224, 224)).to("cuda")
self.ts_model = torch.jit.script(self.model)
@@ -66,6 +68,7 @@
trt_mod = trtorch.embed_engine_in_new_module(trt_engine)
same = (trt_mod(self.input) - self.ts_model(self.input)).abs().max()
self.assertTrue(same < 2e-3)
+
class TestCheckMethodOpSupport(unittest.TestCase):
Reformatting /workspace/tests/modules/hub.py
Reformatting /workspace/tests/py/test_to_backend_api.py
Reformatting /workspace/tests/py/test_api.py
--- /workspace/tests/py/test_trt_intercompatability.py (original)
+++ /workspace/tests/py/test_trt_intercompatability.py (reformatted)
@@ -8,6 +8,7 @@
class TestPyTorchToTRTEngine(ModelTestCase):
+
def setUp(self):
self.input = torch.randn((1, 3, 224, 224)).to("cuda:0")
self.ts_model = torch.jit.script(self.model)
@@ -32,9 +33,12 @@
with engine.create_execution_context() as ctx:
out = torch.empty(size=tuple(engine.get_binding_shape(1))).to("cuda:0")
bindings = [self.input.contiguous().data_ptr(), out.contiguous().data_ptr()]
- ctx.execute_async(batch_size=1, bindings=bindings, stream_handle=torch.cuda.current_stream(device='cuda:0').cuda_stream)
+ ctx.execute_async(batch_size=1,
+ bindings=bindings,
+ stream_handle=torch.cuda.current_stream(device='cuda:0').cuda_stream)
same = (out - self.ts_model(self.input)).abs().max()
self.assertTrue(same < 2e-3)
+
def test_suite():
suite = unittest.TestSuite()
Reformatting /workspace/tests/py/test_ptq_trt_calibrator.py
Reformatting /workspace/tests/py/test_multi_gpu.py
Reformatting /workspace/tests/py/test_api_dla.py
Reformatting /workspace/tests/py/test_trt_intercompatability.py
Reformatting /workspace/tests/py/test_ptq_to_backend.py
Reformatting /workspace/tests/py/model_test_case.py
Reformatting /workspace/tests/py/test_ptq_dataloader_calibrator.py
ERROR: Some files do not conform to style guidelinesSigned-off-by: Naren Dasan <[email protected]> Signed-off-by: Naren Dasan <[email protected]>
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.
Description
This PR adds the ability to register TRT engines from outside TRTorch as TorchScript modules. Modules created through this method inherit all the abilities (and limitations) of TRTorch compiled TorchScript modules including serialization and portability through
torch.jit.loadand deployment withlibtrtorchrt.soFixes #46
Type of change
Please delete options that are not relevant and/or add your own.
Checklist: