[GitHub] [tvm] codeislife99 commented on pull request #7231: Adding aten::unsqueeze_ to PT Frontend
codeislife99 commented on pull request #7231: URL: https://github.com/apache/tvm/pull/7231#issuecomment-759274273 The inplace op recommendation using ` torch._C._jit_pass_remove_mutation(graph) ` didn't work. So I resorted back to the previous option. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 edited a comment on pull request #7231: Adding aten::unsqueeze_ to PT Frontend
codeislife99 edited a comment on pull request #7231: URL: https://github.com/apache/tvm/pull/7231#issuecomment-759255925 I have removed copy_ from this PR and added a few ops after unsqueeze in the test to make sure it works. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] codeislife99 commented on pull request #7231: Adding aten::unsqueeze_ and aten::copy_ ops to PT Frontend
codeislife99 commented on pull request #7231: URL: https://github.com/apache/tvm/pull/7231#issuecomment-759255925 I have removed copy_ from this PR and added a few ops after unsqueeze in the test to make sure it works. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi commented on pull request #7195: [THRUST] Faster multi dimensional argsort by segmented sort
masahi commented on pull request #7195: URL: https://github.com/apache/tvm/pull/7195#issuecomment-759241934 Thanks @mbrookhart @trevor-m This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated: [THRUST] Faster multi dimensional argsort by segmented sort (#7195)
This is an automated email from the ASF dual-hosted git repository. masahi pushed a commit to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git The following commit(s) were added to refs/heads/main by this push: new 1d07f1a [THRUST] Faster multi dimensional argsort by segmented sort (#7195) 1d07f1a is described below commit 1d07f1a0f4e70872c2a52531b6bd8580d64c7538 Author: masahi AuthorDate: Wed Jan 13 15:42:09 2021 +0900 [THRUST] Faster multi dimensional argsort by segmented sort (#7195) * remove sort nms * add segmented sort by key impl * bug fix, test pass * updated fast path condition to work for all dims --- python/tvm/topi/cuda/nms.py | 6 +- python/tvm/topi/cuda/sort.py | 73 +- src/runtime/contrib/thrust/thrust.cu | 117 --- 3 files changed, 72 insertions(+), 124 deletions(-) diff --git a/python/tvm/topi/cuda/nms.py b/python/tvm/topi/cuda/nms.py index 8946446..a4080e5 100644 --- a/python/tvm/topi/cuda/nms.py +++ b/python/tvm/topi/cuda/nms.py @@ -819,11 +819,9 @@ def non_max_suppression( if ( target and target.kind.name == "cuda" -and tvm.get_global_func("tvm.contrib.thrust.sort_nms", allow_missing=True) +and tvm.get_global_func("tvm.contrib.thrust.sort", allow_missing=True) ): -sort_tensor = argsort_thrust( -score_tensor, valid_count=None, axis=1, is_ascend=False, dtype=valid_count_dtype -) +sort_tensor = argsort_thrust(score_tensor, axis=1, is_ascend=False, dtype=valid_count_dtype) else: sort_tensor = argsort(score_tensor, axis=1, is_ascend=False, dtype=valid_count_dtype) diff --git a/python/tvm/topi/cuda/sort.py b/python/tvm/topi/cuda/sort.py index 18872a2..9b6a18a 100644 --- a/python/tvm/topi/cuda/sort.py +++ b/python/tvm/topi/cuda/sort.py @@ -409,68 +409,6 @@ def sort_by_key_ir( ) -def argsort_nms_thrust(data, valid_count, axis=-1, is_ascend=1, dtype="float32"): -"""Performs sorting along the given axis and returns an array of indicies -having same shape as an input array that index data in sorted order. - -Parameters --- -data: tvm.te.Tensor -The input array. - -valid_count : tvm.te.Tensor, optional -The number of valid elements to be sorted. - -axis : int, optional -Axis long which to sort the input tensor. - -is_ascend : boolean, optional -Whether to sort in ascending or descending order. - -dtype : string, optional -DType of the output indices. - -Returns ---- -out : tvm.te.Tensor -The output of this function. -""" -ndim = len(data.shape) -if axis < 0: -axis = ndim + axis -if axis != ndim - 1: -# Prepare for sorting along axis -1. -axes = swap(list(range(ndim)), axis) -data = transpose(data, axes) - -data_buf = tvm.tir.decl_buffer(data.shape, data.dtype, "data_buf", data_alignment=8) -valid_count_buf = tvm.tir.decl_buffer( -valid_count.shape, valid_count.dtype, "valid_count_buf", data_alignment=4 -) -out_bufs = [ -tvm.tir.decl_buffer(data.shape, data.dtype, "value_buf", data_alignment=8), -tvm.tir.decl_buffer(data.shape, "int32", "indices_buf", data_alignment=8), -] -out = te.extern( -[data.shape, data.shape], -[data, valid_count], -lambda ins, outs: tvm.tir.call_packed( -"tvm.contrib.thrust.sort_nms", ins[0], ins[1], outs[0], outs[1], is_ascend -), -in_buffers=[data_buf, valid_count_buf], -out_buffers=out_bufs, -dtype=[data.dtype, "int32"], -name="nms_argsort_gpu", -tag="nms_argsort_gpu", -) - -if axis != ndim - 1: -axes = swap(list(range(ndim)), axis) -out = [transpose(o, axes) for o in out] - -return out[1] - - def sort(data, axis=-1, is_ascend=1): """Performs sorting along the given axis and returns an array of sorted values with the same shape as the input data. @@ -602,7 +540,7 @@ def argsort(data, axis=-1, is_ascend=1, dtype="float32"): return out -def argsort_thrust(data, valid_count=None, axis=-1, is_ascend=1, dtype="float32"): +def argsort_thrust(data, axis=-1, is_ascend=1, dtype="float32"): """Performs sorting along the given axis and returns an array of indicies having same shape as an input array that index data in sorted order. @@ -611,9 +549,6 @@ def argsort_thrust(data, valid_count=None, axis=-1, is_ascend=1, dtype="float32" data: tvm.te.Tensor The input array. -valid_count : tvm.te.Tensor, optional -The number of valid elements to be sorted. - axis : int, optional Axis long which to sort the input tensor. @@ -628,11 +563,7 @@ def argsort_thrust(data, valid_count=None, axis=-1, is_ascend=1, dtype="float32" out : tvm.te.Tensor
[GitHub] [tvm] masahi merged pull request #7195: [THRUST] Faster multi dimensional argsort by segmented sort
masahi merged pull request #7195: URL: https://github.com/apache/tvm/pull/7195 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] wejoncy closed issue #7247: [BUG]( OPENCL Code-generate) Generate float type of index for local memory
wejoncy closed issue #7247: URL: https://github.com/apache/tvm/issues/7247 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] ANSHUMAN87 opened a new pull request #7267: [Frontend][Tensorflow] Sparse dense matmul adjoint option added
ANSHUMAN87 opened a new pull request #7267: URL: https://github.com/apache/tvm/pull/7267 This is a follow up PR! Adjoint option support for both input added in this PR This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] aaltonenzhang commented on issue #7258: tvm doesn't support mix-precision inputs for qnn conv2d
aaltonenzhang commented on issue #7258: URL: https://github.com/apache/tvm/issues/7258#issuecomment-759146686 My account for discussion is on hold, I won’t be able to reply or create topics until a staff member review the status. Could you please give a help? Thanks. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (b5a7de8 -> 86479ba)
This is an automated email from the ASF dual-hosted git repository. zhic pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from b5a7de8 Remove check_correctness in AutoTVM, which is busted (#7250) add 86479ba [Torch] Restore class-aware NMS for detection models by graph rewrite (#7154) No new revisions were added by this update. Summary of changes: python/tvm/relay/frontend/pytorch.py | 14 +- python/tvm/relay/frontend/pytorch_utils.py | 153 - .../frontend/pytorch/test_object_detection.py | 20 ++- 3 files changed, 176 insertions(+), 11 deletions(-)
[GitHub] [tvm] zhiics merged pull request #7154: [Torch] Restore class-aware NMS for detection models by graph rewrite
zhiics merged pull request #7154: URL: https://github.com/apache/tvm/pull/7154 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] areusch opened a new pull request #7266: [µTVM] Add TVMPlatformGenerateRandom, a non-cryptographic random number generator.
areusch opened a new pull request #7266: URL: https://github.com/apache/tvm/pull/7266 * This change is preparation to support autotuning in microTVM. It also cleans up a loose end in the microTVM RPC server implementation. * Randomness is needed in two places of the CRT: 1. to initialize the Session nonce, which provides a more robust way to detect reboots and ensure that messages are not confused across them. 2. to fill input tensors when timing AutoTVM operators (once AutoTVM support lands in the next PR). * This change adds TVMPlatformGenerateRandom, a platform function for generating non-cryptographic random data, to service those needs. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] ANSHUMAN87 commented on pull request #7148: [Frontend][Tensorflow] Sparse_Dense Op CSR scheduling issue resolved for Cuda & X86
ANSHUMAN87 commented on pull request #7148: URL: https://github.com/apache/tvm/pull/7148#issuecomment-759136347 Gentle ping @tkonolige ! cc @junrushao1994 , @comaniac too :) This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] ANSHUMAN87 commented on pull request #7048: [Frontend][TFLite] Densify Op added
ANSHUMAN87 commented on pull request #7048: URL: https://github.com/apache/tvm/pull/7048#issuecomment-759135312 Gentle ping @zhiics , @FrozenGene ! This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] areusch commented on pull request #7250: Remove check_correctness in AutoTVM, which is busted
areusch commented on pull request #7250: URL: https://github.com/apache/tvm/pull/7250#issuecomment-759114856 discussed with @antinucleon and we think it should be okay to just remove this. if anyone is using this, please comment back on this thread and we'll consider a way forward. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (4364386 -> b5a7de8)
This is an automated email from the ASF dual-hosted git repository. tqchen pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from 4364386 Add op_name in error message for Pool (#7243) add b5a7de8 Remove check_correctness in AutoTVM, which is busted (#7250) No new revisions were added by this update. Summary of changes: python/tvm/autotvm/measure/measure_methods.py | 60 -- tests/python/unittest/test_autotvm_measure.py | 28 -- vta/scripts/tune_conv2d.py | 2 +- vta/scripts/tune_conv2d_transpose.py | 2 +- vta/scripts/tune_dense.py | 2 +- vta/scripts/tune_group_conv2d.py | 2 +- vta/scripts/tune_resnet.py | 2 +- .../python/integration/test_benchmark_gemm.py | 23 - vta/tutorials/autotvm/tune_relay_vta.py| 2 +- 9 files changed, 26 insertions(+), 97 deletions(-)
[GitHub] [tvm] tqchen merged pull request #7250: Remove check_correctness in AutoTVM, which is busted
tqchen merged pull request #7250: URL: https://github.com/apache/tvm/pull/7250 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (ac684f9 -> 4364386)
This is an automated email from the ASF dual-hosted git repository. ziheng pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from ac684f9 Fix TRT weight conversion when first dim of weight shape is 1 (#7253) add 4364386 Add op_name in error message for Pool (#7243) No new revisions were added by this update. Summary of changes: python/tvm/relay/frontend/onnx.py | 16 1 file changed, 8 insertions(+), 8 deletions(-)
[GitHub] [tvm] ZihengJiang commented on pull request #7243: Add op_name in error message for Pool
ZihengJiang commented on pull request #7243: URL: https://github.com/apache/tvm/pull/7243#issuecomment-759108993 Merged. Thanks @luyaor @jwfromm This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] ZihengJiang merged pull request #7243: Add op_name in error message for Pool
ZihengJiang merged pull request #7243: URL: https://github.com/apache/tvm/pull/7243 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (e3b2984 -> ac684f9)
This is an automated email from the ASF dual-hosted git repository. comaniac pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from e3b2984 Do not use ICHECK in nnvm (#7255) add ac684f9 Fix TRT weight conversion when first dim of weight shape is 1 (#7253) No new revisions were added by this update. Summary of changes: python/tvm/relay/op/contrib/tensorrt.py | 6 +- src/runtime/contrib/tensorrt/tensorrt_builder.cc | 18 -- tests/python/contrib/test_tensorrt.py| 2 ++ 3 files changed, 19 insertions(+), 7 deletions(-)
[GitHub] [tvm] comaniac commented on pull request #7253: [BYOC][TRT] Fix weight conversion when first dim of weight shape is 1
comaniac commented on pull request #7253: URL: https://github.com/apache/tvm/pull/7253#issuecomment-759100534 Thanks @trevor-m This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] comaniac merged pull request #7253: [BYOC][TRT] Fix weight conversion when first dim of weight shape is 1
comaniac merged pull request #7253: URL: https://github.com/apache/tvm/pull/7253 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] sxjscience commented on a change in pull request #7230: [FRONTEND][Mxnet][nuymp] Adding _npi_advanced_indexing_multiple
sxjscience commented on a change in pull request #7230: URL: https://github.com/apache/tvm/pull/7230#discussion_r556152281 ## File path: tests/python/frontend/mxnet/test_forward.py ## @@ -1935,6 +1935,29 @@ def verify(data_shape, axis, use_length, length): verify((2, 3, 4), 2, True, np.array([[3, 4, 2], [1, 2, 1]]).astype("int32")) +@pytest.mark.parametrize( +"data_shape, row_sel, col", +[ +((5, 7), (0, 1, 2, 3, 4,), 2), +], +) +@pytest.mark.parametrize("dtype", ["float64", "float32"]) +@tvm.testing.parametrize_targets +@pytest.mark.parametrize("kind", ["graph", "vm", "debug"]) +def test_forward_npi_advanced_indexing_multiple(data_shape, row_sel, col, dtype, target, ctx, kind): +data_np = np.random.uniform(size=data_shape).astype(dtype) +data = mx.sym.var("data") +ref_res = mx.np.array(data_np)[row_sel, col] + +# TODO need to add the proper symbol operator +mx_sym = mx.sym.np.(data.as_np_ndarray()[row_sel, col]) Review comment: Also, I think you will need to install the nightly version of MXNet 2.0. You may follow the guide in https://github.com/dmlc/gluon-nlp. ```bash # Install the version with CUDA 10.1 python3 -m pip install -U --pre "mxnet-cu101>=2.0.0b20201206" -f https://dist.mxnet.io/python # Install the version with CUDA 10.2 python3 -m pip install -U --pre "mxnet-cu102>=2.0.0b20201206" -f https://dist.mxnet.io/python # Install the version with CUDA 11 python3 -m pip install -U --pre "mxnet-cu110>=2.0.0b20201206" -f https://dist.mxnet.io/python # Install the cpu-only version python3 -m pip install -U --pre "mxnet>=2.0.0b20201206" -f https://dist.mxnet.io/python ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] gromero opened a new pull request #7265: [µTVM] Avoid listing links when probing serial ports
gromero opened a new pull request #7265: URL: https://github.com/apache/tvm/pull/7265 SerialTransport.open() probes automatically the device name based upon a grep regex if a device name is not provided. The code expects to find only a single device. Currently when it probes for the available serial ports it includes in the list the device names that are also symbolic links. Since _find_openocd_serial_port() always returns a serial number for a given serial port (not the device name path) the available device names are always probed when the openocd flash runner is used. It's not uncommon that device drivers create symbolic links for certain kinds of serial devices, specially those that provide a serial port plus an additional endpoint to program the device attached, like a ST-Link interface, etc. As a consequence the current code fails to select the correct device name when symbolic links exist and the openocd flash runner is used. That commit changes the probe behavior to avoid listing symbolic links when probing the device name for the target serial port. Without that change the following error happens: ``` Traceback (most recent call last): File "./micro_tflite.py", line 255, in with tvm.micro.Session(binary=micro_binary, flasher=flasher) as session: File "/home/gromero/git/tvm/python/tvm/micro/session.py ", line 127, in __enter__ self.transport = TransportLogger( File "/home/gromero/git/tvm/python/tvm/micro/transport/base.py", line 79, in __enter__ self.open() File "/home/gromero/git/tvm/python/tvm/micro/transport/base.py", line 207, in open self.child.open() File "/home/gromero/git/tvm/python/tvm/micro/transport/serial.py", line 72, in open raise SerialPortNotFoundError( NameError: name 'SerialPortNotFoundError' is not defined When ARM `STM32 STLink` is used by a board and the following device names are created: ``` ``` gromero@gromero0:~/git/tvm$ ls -l /dev/{ttyACM0,stlinkv2-1_0} lrwxrwxrwx 1 root root 7 Jan 12 16:10 /dev/stlinkv2-1_0 -> ttyACM0 crw-rw-rw- 1 root plugdev 166, 0 Jan 12 21:32 /dev/ttyACM0 ``` Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from [Reviewers](https://github.com/apache/incubator-tvm/blob/master/CONTRIBUTORS.md#reviewers) by @ them in the pull request thread. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] rkimball opened a new pull request #7264: Change the all #pragma once to ifdef include guard
rkimball opened a new pull request #7264: URL: https://github.com/apache/tvm/pull/7264 These are all of the instances I could find outside of 3rdparty This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi edited a comment on pull request #7257: [TOPI] Improve memory layout inside GPU NMS kernel
masahi edited a comment on pull request #7257: URL: https://github.com/apache/tvm/pull/7257#issuecomment-758961671 Yes, ideally I want to update our NMS to be closer to TF/ONNX/PyTorch, and let MXNet frontend handle split and concat, rather than the other way around (what we have now). Current API is over complicated due to the need to support both styles. If we can assume that `return_indices` is always True, we can clean up our API a lot. For example, `invalid_to_bottom` argument only makes sense for MXNet. We don't need `coord_start`, `score_index`, and `id_index` arguments, if inputs are only unpacked. Supporting `max_output_boxes_per_class` needs change in the implementation as well. We need to count the number of survived boxes per class. But that's the only change I think, it is definitely doable without writing another kernel. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi commented on pull request #7257: [TOPI] Improve memory layout inside GPU NMS kernel
masahi commented on pull request #7257: URL: https://github.com/apache/tvm/pull/7257#issuecomment-758961671 Yes, ideally I want to update our NMS to be more closer to TF/ONNX/PyTorch, and let MXNet frontend handle split and concat, rather than the other way around (what we have now). Current API is over complicated due to the need to support both styles. If we can assume that `return_indices` is always True, we can clean up our API a lot. For example, `invalid_to_bottom` argument only makes sense for MXNet. We don't need `coord_start`, `score_index`, and `id_index` arguments, if inputs are only unpacked. Supporting `max_output_boxes_per_class` needs change in the implementation as well. We need to count the number of survived boxes per class. But that's the only change I think, it is definitely doable without writing another kernel. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] tqchen commented on issue #6792: [TVMC] TODO items on TVMC backlog
tqchen commented on issue #6792: URL: https://github.com/apache/tvm/issues/6792#issuecomment-758914472 @leandron please update the state of the thread accordingly This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi edited a comment on pull request #7257: [TOPI] Improve memory layout inside GPU NMS kernel
masahi edited a comment on pull request #7257: URL: https://github.com/apache/tvm/pull/7257#issuecomment-758901561 I don't quite follow, maybe you are missing something? First, this PR doesn't change our NMS API, it only changes the buffer layouts used internally. Second, the final concat is only required for MXNet, which uses `return_indices=False`. Our NMS returns something completely different depending on `return_indices` flag : If True, it returns a big output tensor packed with bboxes, scores and class ids, with invalid entries indicated by -1. (The valid entries are supposed to move to the top, if ` invalid_to_bottom` flag is True. But our GPU NMS kernel ignores this argument and the output is not reordered. This is another difference with CPU implementation, I think this is a bug) https://github.com/apache/tvm/blob/4e8cc4fc26e931e38017d198d29f45cba04f5a60/python/tvm/topi/cuda/nms.py#L762-L763 If `return_indices=True`, which applies to TF, ONNX, and PyTorch, we only return survived box indices, so there is no need to concat bboxes, scores, and class ids. For this case, there is zero additional overhead after this PR, it is just that concat before NMS done by frontends are now completely pointless. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi commented on pull request #7257: [TOPI] Improve memory layout inside GPU NMS kernel
masahi commented on pull request #7257: URL: https://github.com/apache/tvm/pull/7257#issuecomment-758901561 I don't quite follow, maybe you are missing something? First, this PR doesn't change our NMS API, it only changes the buffer layouts used internally. Second, the final concat is only required for MXNet, which uses `return_indices=False`. Our NMS returns something completely different depending on `return_indices` flag : If True, it returns a big output tensor packed with bboxes, scores and class ids, with invalid entries indicated by -1. (The valid entries are supposed to move to the top, if ` invalid_to_bottom` flag is True. But our GPU NMS kernel ignores this argument and the output is not reordered. This is another difference with CPU implementation, I think this is a bug) https://github.com/apache/tvm/blob/4e8cc4fc26e931e38017d198d29f45cba04f5a60/python/tvm/topi/cuda/nms.py#L762-L763 If `return_indices=True`, which applies to TF, ONNX, and PyTorch, we only return survived box indices, so there is no need to concat bboxes, scores, and class ids. For this case, there is zero additional overhead, it is just that concat before NMS done by frontends are now completely pointless. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] leandron commented on issue #7261: [BUG][BYOC][Ethos-N]unable to compile tvm with ethos-n-driver-stack
leandron commented on issue #7261: URL: https://github.com/apache/tvm/issues/7261#issuecomment-758859469 Hi @gaintpd, Have you built the driver stack, prior to building TVM? If not, you can follow the [official documentation](https://github.com/ARM-software/ethos-n-driver-stack#build-tools), or the steps described here https://github.com/apache/tvm/blob/main/docker/install/ubuntu_install_ethosn_driver_stack.sh. Hope that helps. Also cc @Leo-arm. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] comaniac commented on issue #7261: [BUG][BYOC][Ethos-N]unable to compile tvm with ethos-n-driver-stack
comaniac commented on issue #7261: URL: https://github.com/apache/tvm/issues/7261#issuecomment-758857442 cc @mbaret This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] mbrookhart commented on pull request #7257: [TOPI] Improve memory layout inside GPU NMS kernel
mbrookhart commented on pull request #7257: URL: https://github.com/apache/tvm/pull/7257#issuecomment-758777131 Even after this, I think we will still need a loop over classes for ONNX and TF, since ONNX explicitly and TF implicitly need max_output_boxes_per_class, while this op even with class id will return max_output_boxes for all classes. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] tkonolige commented on pull request #7235: [FIX] Fix make format to work with arbitrary upstream names
tkonolige commented on pull request #7235: URL: https://github.com/apache/tvm/pull/7235#issuecomment-758775546 I changed `make format` back to use just upstream/main. We were also using origin/main in some other places, which I've switched to upstream/main. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] luyaor commented on pull request #7243: Add op_name in error message for Pool
luyaor commented on pull request #7243: URL: https://github.com/apache/tvm/pull/7243#issuecomment-758725279 @jwfromm Hi, could you help to merge this PR? Or anyone else is needed for approving review? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] luyaor opened a new issue #7263: [Bug] [Frontend] Matrix C of Gemm operator in ONNX could be optional
luyaor opened a new issue #7263: URL: https://github.com/apache/tvm/issues/7263 ## Description When compiling following model with TVM, it will error on parsing. I think it is because matrix C for inputs could be optional by ONNX specification. Check here: https://github.com/onnx/onnx/blob/master/docs/Operators.md#gemm The model(with ONNX as frontend) with error is as follows, check bug.onnx in [bug7.zip](https://github.com/apache/tvm/files/5802949/bug7.zip). ## Error Log ``` Traceback (most recent call last): File "check.py", line 11, in mod, params = relay.frontend.from_onnx(onnx_model, {}) File "/Users/luyaor/Documents/tvm/python/tvm/relay/frontend/onnx.py", line 2806, in from_onnx mod, params = g.from_onnx(graph, opset, freeze_params) File "/Users/luyaor/Documents/tvm/python/tvm/relay/frontend/onnx.py", line 2613, in from_onnx op = self._convert_operator(op_name, inputs, attr, opset) File "/Users/luyaor/Documents/tvm/python/tvm/relay/frontend/onnx.py", line 2721, in _convert_operator sym = convert_map[op_name](inputs, attrs, self._params) File "/Users/luyaor/Documents/tvm/python/tvm/relay/frontend/onnx.py", line 510, in _impl_v1 assert len(inputs) == 3, "Gemm op take 3 inputs, {} given".format(len(inputs)) AssertionError: Gemm op take 3 inputs, 2 given ``` ## How to reproduce ### Environment Python3, with tvm, onnx tvm version: [`c31e338`](https://github.com/apache/tvm/commit/c31e338d5f98a8e8c97286c5b93b20caee8be602) Wed Dec 9 14:52:58 2020 +0900 1. Download [bug7.zip](https://github.com/apache/tvm/files/5802949/bug7.zip) 2. Run `python check.py`. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] luyaor opened a new issue #7262: [Bug] Error when compiling a ONNX model with Gemm operator
luyaor opened a new issue #7262: URL: https://github.com/apache/tvm/issues/7262 ## Description When compiling following model with TVM, it will crash. The model(with ONNX as frontend) with error is as follows, check bug.onnx in [bug6.zip](https://github.com/apache/tvm/files/5802832/bug6.zip). ![image](https://user-images.githubusercontent.com/7541296/104329942-7d0d2b80-5528-11eb-83dc-ac3f77e30936.png) The corresponding relay program: ``` #[version = "0.0.5"] def @main(%node1: Tensor[(4, 8), float32], %node2: Tensor[(8, 5), float32], %node3: Tensor[(4, 5), float32]) { %0 = nn.batch_flatten(%node1); %1 = transpose(%node2, axes=[1, 0]); %2 = nn.dense(%0, %1, units=5); %3 = multiply(1f, %node3); nn.bias_add(%2, %3) } ``` ## Error Log ``` tensor type `Tensor[(5), float32]` has 1 dimensions, while `Tensor[(4, 5), float32]` has 2 dimensions The Relay type checker is unable to show the following types match. In particular `Tensor[(5), float32]` does not match `Tensor[(4, 5), float32]` Traceback (most recent call last): File "check.py", line 19, in tvm_graph, tvm_lib, tvm_params = relay.build_module.build(mod, target, params=params) File "/Users/luyaor/Documents/tvm/python/tvm/relay/build_module.py", line 275, in build graph_json, mod, params = bld_mod.build(mod, target, target_host, params) File "/Users/luyaor/Documents/tvm/python/tvm/relay/build_module.py", line 138, in build self._build(mod, target, target_host) File "/Users/luyaor/Documents/tvm/python/tvm/_ffi/_ctypes/packed_func.py", line 237, in __call__ raise get_last_ffi_error() tvm.error.DiagnosticError: Traceback (most recent call last): [bt] (8) 9 libtvm.dylib0x00011a18ab7c tvm::transform::PassNode::operator()(tvm::IRModule) const + 60 [bt] (7) 8 libtvm.dylib0x00011a286415 tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const + 869 [bt] (6) 7 libtvm.dylib0x00011a286792 tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const + 194 [bt] (5) 6 libtvm.dylib0x00011a2862fc tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const + 588 [bt] (4) 5 libtvm.dylib0x00011a286792 tvm::transform::Pass::operator()(tvm::IRModule, tvm::transform::PassContext const&) const + 194 [bt] (3) 4 libtvm.dylib0x00011a284f25 tvm::transform::ModulePassNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const + 789 [bt] (2) 3 libtvm.dylib0x00011acc53f9 std::__1::__function::__func::AssignTypedLambda(tvm::relay::transform::InferType()::$_1)::'lambda'(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*), std::__1::allocator::AssignTypedLambda(tvm::relay::transform::InferType()::$_1)::'lambda'(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*)>, void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)>::operator()(tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&) + 2025 [bt] (1) 2 libtvm.dylib0x00011a2440cc tvm::DiagnosticContext::Render() + 476 [bt] (0) 1 libtvm.dylib0x00011a017c6f dmlc::LogMessageFatal::~LogMessageFatal() + 111 File "/Users/luyaor/Documents/tvm/src/ir/diagnostic.cc", line 105 DiagnosticError: one or more error diagnostics were emitted, please check diagnostic render for output. ``` ## How to reproduce ### Environment Python3, with tvm, onnx tvm version: [`c31e338`](https://github.com/apache/tvm/commit/c31e338d5f98a8e8c97286c5b93b20caee8be602) Wed Dec 9 14:52:58 2020 +0900 1. Download [bug6.zip](https://github.com/apache/tvm/files/5802832/bug6.zip) 2. Run `python check.py`. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] luyaor commented on issue #7244: [Bug] [Relay] Error when compiling ONNX with LeakyRelu
luyaor commented on issue #7244: URL: https://github.com/apache/tvm/issues/7244#issuecomment-758703453 Fixed in #7259. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] luyaor closed issue #7244: [Bug] [Relay] Error when compiling ONNX with LeakyRelu
luyaor closed issue #7244: URL: https://github.com/apache/tvm/issues/7244 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] tqchen commented on issue #7258: tvm doesn't support mix-precision inputs for qnn conv2d
tqchen commented on issue #7258: URL: https://github.com/apache/tvm/issues/7258#issuecomment-758676410 This seems to be a discussion for enhancement, would be great to open a thread on https://discuss.tvm.apache.org/ This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] tqchen closed issue #7258: tvm doesn't support mix-precision inputs for qnn conv2d
tqchen closed issue #7258: URL: https://github.com/apache/tvm/issues/7258 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] tqchen commented on pull request #7235: [FIX] Fix make format to work with arbitrary upstream names
tqchen commented on pull request #7235: URL: https://github.com/apache/tvm/pull/7235#issuecomment-758666080 I see, given that we are encourage a linear chain history, and most PRs expect rebasing to the latest main, I would imagine this would not be an issue. I would still recommend go with the `upstream/main` recommendation since it is simpler, and developer can swap with other possible git base commit, without having to introducing another layer of indirection. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (b84eb16 -> e3b2984)
This is an automated email from the ASF dual-hosted git repository. tqchen pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from b84eb16 [ONNX] Fix issues for Clip and RoiAlign (#7237) add e3b2984 Do not use ICHECK in nnvm (#7255) No new revisions were added by this update. Summary of changes: nnvm/include/nnvm/graph.h | 4 ++-- nnvm/include/nnvm/layout.h| 40 +++ nnvm/include/nnvm/op.h| 12 ++-- nnvm/include/nnvm/tuple.h | 4 ++-- nnvm/src/core/graph.cc| 10 +- nnvm/src/core/op.cc | 2 +- nnvm/src/core/pass.cc | 2 +- nnvm/src/core/symbolic.cc | 22 ++--- nnvm/src/pass/correct_layout.cc | 12 ++-- nnvm/src/pass/gradient.cc | 16 nnvm/src/pass/graph_algorithm.h | 10 +- nnvm/src/pass/infer_shape_type.cc | 24 +++ nnvm/src/pass/place_device.cc | 12 ++-- nnvm/src/pass/plan_memory.cc | 4 ++-- nnvm/src/pass/print_graph_ir.cc | 2 +- nnvm/src/pass/saveload_json.cc| 18 +- nnvm/tests/cpp/op_test.cc | 2 +- nnvm/tests/cpp/tuple_test.cc | 8 18 files changed, 102 insertions(+), 102 deletions(-)
[GitHub] [tvm] tqchen merged pull request #7255: Do not use ICHECK in nnvm
tqchen merged pull request #7255: URL: https://github.com/apache/tvm/pull/7255 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] gaintpd opened a new issue #7261: [BUG][BYOC][Ethos-N]unable to compile tvm with ethos-n-driver-stack
gaintpd opened a new issue #7261: URL: https://github.com/apache/tvm/issues/7261 I can compile the tvm successfully on my ubuntu 20.04 LTS edition, and the hardware is thinkpad P52, When I set the ethos-n option in config.cmake in the build directory, I get the following error: ``` -- Build with RPC support... -- Build with Graph runtime support... -- Build with Graph runtime debug support... -- VTA build with VTA_HW_PATH=/home/dyn/tvm/3rdparty/vta-hw -- Build VTA runtime with target: sim -- Found CUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda -- Found CUDA_CUDA_LIBRARY=/usr/local/cuda/targets/x86_64-linux/lib/stubs/libcuda.so -- Found CUDA_CUDART_LIBRARY=/usr/local/cuda/lib64/libcudart.so -- Found CUDA_NVRTC_LIBRARY=/usr/local/cuda/lib64/libnvrtc.so -- Found CUDA_CUDNN_LIBRARY=CUDA_CUDNN_LIBRARY-NOTFOUND -- Found CUDA_CUBLAS_LIBRARY=/usr/local/cuda/lib64/libcublas.so -- Found CUDA_CUBLASLT_LIBRARY=/usr/local/cuda/lib64/libcublasLt.so -- Build with CUDA support -- Not found - LLVM_LIBS -- Fall back to using llvm-config -- Use llvm-config=/usr/lib/llvm-10/bin/llvm-config -- Found LLVM_INCLUDE_DIRS=/usr/lib/llvm-10/include -- Found LLVM_DEFINITIONS=-D_GNU_SOURCE -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS;-D_GNU_SOURCE;-D__STDC_CONSTANT_MACROS;-D__STDC_FORMAT_MACROS;-D__STDC_LIMIT_MACROS -- Found LLVM_LIBS=/usr/lib/llvm-10/lib/libLLVM-10.so -- Found TVM_LLVM_VERSION=100 -- Build with LLVM 10.0.0 -- Set TVM_LLVM_VERSION=100 -- Arm Ethos-N driver stack PATH=/media/dyn/Data/deepLearningLibrary-linux/ethos-n-driver-stack CMake Error at cmake/modules/contrib/EthosN.cmake:24 (message): Cannot find Ethos-N, USE_ETHOSN=/media/dyn/Data/deepLearningLibrary-linux/ethos-n-driver-stack Call Stack (most recent call first): CMakeLists.txt:339 (include) -- Configuring incomplete, errors occurred! See also "/home/dyn/tvm/build/CMakeFiles/CMakeOutput.log". See also "/home/dyn/tvm/build/CMakeFiles/CMakeError.log". ``` How to fix this problem? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] wejoncy commented on issue #7247: [BUG]( OPENCL Code-generate) Generate float type of index for local memory
wejoncy commented on issue #7247: URL: https://github.com/apache/tvm/issues/7247#issuecomment-758630626 Have root cause the issue, `te.thread_axis` do not check if dom type is int. So add a cast operation in `te.thread_axis` could resolve it. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] jcf94 opened a new pull request #7260: [AutoScheduler] Bug fix & Custom sketch support
jcf94 opened a new pull request #7260: URL: https://github.com/apache/tvm/pull/7260 - Some bug fix for cost model modifications in #7197 - Finally we bring the custom sketch support cc @comaniac @merrymercy This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] zhanghaohit commented on pull request #6126: [VTA][OpenCL] intelfocl
zhanghaohit commented on pull request #6126: URL: https://github.com/apache/tvm/pull/6126#issuecomment-758592568 @tmoreau89 CI seems not working? wait for 9 hours but not start yet. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] insop commented on a change in pull request #7230: [FRONTEND][Mxnet][nuymp] Adding _npi_advanced_indexing_multiple
insop commented on a change in pull request #7230: URL: https://github.com/apache/tvm/pull/7230#discussion_r555663382 ## File path: tests/python/frontend/mxnet/test_forward.py ## @@ -1935,6 +1935,29 @@ def verify(data_shape, axis, use_length, length): verify((2, 3, 4), 2, True, np.array([[3, 4, 2], [1, 2, 1]]).astype("int32")) +@pytest.mark.parametrize( +"data_shape, row_sel, col", +[ +((5, 7), (0, 1, 2, 3, 4,), 2), +], +) +@pytest.mark.parametrize("dtype", ["float64", "float32"]) +@tvm.testing.parametrize_targets +@pytest.mark.parametrize("kind", ["graph", "vm", "debug"]) +def test_forward_npi_advanced_indexing_multiple(data_shape, row_sel, col, dtype, target, ctx, kind): +data_np = np.random.uniform(size=data_shape).astype(dtype) +data = mx.sym.var("data") +ref_res = mx.np.array(data_np)[row_sel, col] + +# TODO need to add the proper symbol operator +mx_sym = mx.sym.np.(data.as_np_ndarray()[row_sel, col]) Review comment: Hi @sxjscience Thank you for the suggestion. I think I had tried that as well. So with the following, I got this exception `IndexError: Only integer, slice, or tuple of these types are supported! Received key=(<_Symbol row_sel>, <_Symbol col>)` from `mxnet/symbol/numpy/_symbol.py:135:` ([link](https://github.com/apache/incubator-mxnet/blob/124d8417984ed9205f972eb1c2cadbf028b94eb3/python/mxnet/symbol/numpy/_symbol.py#L149)) I will dig more, but if you have any suggestion, please let me know. ``` def test_forward_npi_advanced_indexing_multiple(data_shape, row_sel, col, dtype, target, ctx, kind): data_np = np.random.uniform(size=data_shape).astype(dtype) ref_res = mx.np.array(data_np)[row_sel, col] row_sel_sym = mx.sym.var("row_sel").as_np_ndarray() data_sym = mx.sym.var("data").as_np_ndarray() col_sym = mx.sym.var("col").as_np_ndarray() mx_sym = data_sym[row_sel_sym, col_sym] mod, _ = relay.frontend.from_mxnet( mx_sym, shape={"data": data_shape, "row_sel": row_sel, "col": col}, dtype=dtype ) intrp = relay.create_executor(kind, mod=mod, ctx=ctx, target=target) op_res = intrp.evaluate()(data_np) tvm.testing.assert_allclose(op_res.asnumpy(), ref_res.asnumpy(), rtol=1e-5) tvm.testing.assert_allclose(ref_res.asnumpy(), ref_res.asnumpy(), rtol=1e-5) ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (10b7929 -> b84eb16)
This is an automated email from the ASF dual-hosted git repository. masahi pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from 10b7929 add default value for leaky relu alpha (#7259) add b84eb16 [ONNX] Fix issues for Clip and RoiAlign (#7237) No new revisions were added by this update. Summary of changes: python/tvm/relay/frontend/onnx.py | 8 ++-- tests/python/frontend/onnx/test_forward.py | 24 ++-- 2 files changed, 28 insertions(+), 4 deletions(-)
[GitHub] [tvm] masahi merged pull request #7237: [ONNX] Fix issues for Clip and RoiAlign
masahi merged pull request #7237: URL: https://github.com/apache/tvm/pull/7237 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[tvm] branch main updated (72c9a51 -> 10b7929)
This is an automated email from the ASF dual-hosted git repository. masahi pushed a change to branch main in repository https://gitbox.apache.org/repos/asf/tvm.git. from 72c9a51 [FIX,TUTORIALS] Import tvm.testing in tutorials that use it (#7248) add 10b7929 add default value for leaky relu alpha (#7259) No new revisions were added by this update. Summary of changes: python/tvm/relay/op/nn/nn.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
[GitHub] [tvm] masahi merged pull request #7259: [ONNX Frontend] add default value for leaky relu alpha
masahi merged pull request #7259: URL: https://github.com/apache/tvm/pull/7259 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi edited a comment on pull request #7154: [Torch] Restore class-aware NMS for detection models by graph rewrite
masahi edited a comment on pull request #7154: URL: https://github.com/apache/tvm/pull/7154#issuecomment-758550210 @kevinthesun @zhiics @mbrookhart As shown in my new NMS PR https://github.com/apache/tvm/pull/7257, this rewrite results in a better speed up with improved memory layout. Can we merge this? I have new rewrites coming to further optimize PyTorch NMS and MaskRCNN / FasterRCNN. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] masahi commented on pull request #7154: [Torch] Restore class-aware NMS for detection models by graph rewrite
masahi commented on pull request #7154: URL: https://github.com/apache/tvm/pull/7154#issuecomment-758550210 @kevinthesun @zhiics @mbrookhart As shown in my new NMS PR https://github.com/apache/tvm/pull/7257, this rewrite results in a better speed up with improved memory layout. Can we merge this? I have newer rewrites coming to further optimize PyTorch NMS and MaskRCNN / FasterRCNN This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
[GitHub] [tvm] lixiaoquan edited a comment on pull request #7237: [ONNX] Fix issues for Clip and RoiAlign
lixiaoquan edited a comment on pull request #7237: URL: https://github.com/apache/tvm/pull/7237#issuecomment-758478199 > Maybe you can tell us what the issue was. For RoiAlign, 1) attr.get(“mode", "avg”) will return a string "avg" as default value, which always fails the following test `if mode == b"avg"` 2) infer_type(rois).type_annotation only works when rois is variable, but it could be CallNode For Clip 1) It is possible "min"/"max" don't exist in attrs This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org