[RELAY][MXNET][FRONTEND] add support for MXNET numpy operators #6054

sandyhu533 · 2020-07-14T13:36:54Z

add support for some MXNET numpy operators, related issue: dmlc/gluon-nlp#1244
For now, these operators have been implemented

_npi_transpose
_npi_pad
_npi_multiply_scalar
_npi_true_divide_scalar
_npi_add
_npi_concatenate
_npi_multiply
_np_copy
_npi_tanh
_npi_power_scalar
_npi_less
_npi_add_scalar
_npx_reshape
_split_v2
_npi_where_rscalar

sxjscience · 2020-07-15T19:19:48Z

tests/python/frontend/mxnet/test_forward.py

+
+    def verify(data_shape, scalar):
+        cond_np = np.random.uniform(size=data_shape).astype("bool")
+        data_np = np.random.uniform(size=data_shape).astype("float32")


We may need to test for multiple dtypes. Especially for cond_np, we can have bool or float32.

yzhliu · 2020-07-17T20:13:01Z

tests/python/frontend/mxnet/test_forward.py

@@ -1350,7 +1351,7 @@ def verify(batch, seq_length, num_heads, head_dim):
    verify(1, 10, 4, 16)
    verify(3, 10, 6, 8)

-
+    


remove the change

yzhliu · 2020-07-17T20:16:46Z

tests/python/frontend/mxnet/test_forward.py

+        pytest.skip("mx.sym.np.pad hasn't been publish yet")
+
+    def verify(data_shape, mode, pad_width, constant_value=0.0):
+        for dtype in dtype_list:


would it be easier to use @pytest.mark.parametrize?

yzhliu · 2020-07-17T20:17:33Z

tests/python/frontend/mxnet/test_forward.py

+    test_forward_npi_binary_scalar()
+    test_forward_npi_tanh()
+    test_forward_npi_where_rscalar()
+    test_forward_split_v2()


maybe we can just do pytest.main([__file__])

yzhliu · 2020-07-17T20:17:58Z

tests/python/frontend/mxnet/test_forward.py

+
+def test_forward_npi_pad():
+    if not hasattr(mx.sym.np, 'pad'):
+        pytest.skip("mx.sym.np.pad hasn't been publish yet")


do we need to update CI's mxnet version?

I'm afraid we couldn't update CI's mxnet version now. The latest version of mxnet has made some changes that the test functions in tvm didn't updated with them. So the CI would fail if we update mxnet version now.
This is the error message when I try to run pytest for tests/python/frontend/mxnet/test_forward.py.
E AttributeError: module 'mxnet' has no attribute 'mod'

We will need to find a good way to support both mx 1.x and 2.0.

For marking the test, maybe use @pytest.mark.skipif ?

Why not use two separate pipelines to test with the different mxnet versions?

@tqchen what do you think?

skip ci for now sounds good

tqchen · 2020-07-21T15:22:54Z

cc @antinucleon @junrushao1994

yzhliu · 2020-07-28T00:57:26Z

tests/python/frontend/mxnet/test_forward.py

+        mx_sym = mx.sym.np.pad(data.as_np_ndarray(), mode=mode, pad_width=pad_width)
+    mod, _ = relay.frontend.from_mxnet(mx_sym, {"data": data_shape}, dtype=dtype)
+    for target, ctx in ctx_list():
+        for kind in ["debug"]:


why is this one using debug only?

maybe we can also use @pytest.mark.parametrize for ctx_list and kind.

yzhliu · 2020-07-28T01:01:08Z

tests/python/frontend/mxnet/test_forward.py

+    verify((3, 2, 1), axis=0, indices_or_sections=3, squeeze_axis=True)
+    verify((3, 2, 1), axis=0, indices_or_sections=(1, 2))
+
+
 if __name__ == '__main__':
    test_forward_mlp()


remove all the test_xxx function calls and add pytest.main([__file__])

yzhliu · 2020-07-30T03:55:16Z

Please check the ci failure, otherwise good to me.

…-tvm into relay_mx_np

yzhliu · 2020-08-21T20:48:09Z

Thanks @sandyhu533 @sxjscience

sxjscience · 2020-08-21T20:49:08Z

Thanks! Cannot wait to try it in GluonNLP 👍

…e#6054) * [RELAY][MXNET][FRONTEND] add supports for OPs in numpy from mxnet * Update test_forward.py * Update mxnet.py * Update mxnet.py * Update test_forward.py * update and bugfix * test for multiple dtypes * Update test_forward.py * add data type and optimize coding style * replace pytest.skip with @pytest.mark.skipif * Update test_forward.py * update pytest style * Update test_forward.py * Update test_forward.py * Update test_forward.py * Update test_forward.py Co-authored-by: Ubuntu <ubuntu@ip-172-31-39-169.ap-northeast-1.compute.internal>

Ubuntu and others added 7 commits July 14, 2020 13:25

[RELAY][MXNET][FRONTEND] add supports for OPs in numpy from mxnet

b04dad4

Merge branch 'master' into relay_mx_np

7f1e329

Update test_forward.py

851f47a

Update mxnet.py

fb1d50b

Update mxnet.py

5188efa

Update test_forward.py

781dd73

update and bugfix

45fdde4

sxjscience reviewed Jul 15, 2020

View reviewed changes

sandyhu533 added 2 commits July 16, 2020 18:27

test for multiple dtypes

e8f25bc

Update test_forward.py

bca8e36

yzhliu reviewed Jul 17, 2020

View reviewed changes

sandyhu533 force-pushed the relay_mx_np branch from b326744 to bca8e36 Compare July 20, 2020 14:35

add data type and optimize coding style

04e5552

tqchen assigned yzhliu Jul 21, 2020

sandyhu533 added 3 commits July 22, 2020 23:23

replace pytest.skip with @pytest.mark.skipif

84ab57e

Merge branch 'master' into relay_mx_np

31630ad

Update test_forward.py

9151019

yzhliu reviewed Jul 28, 2020

View reviewed changes

update pytest style

cb08135

sandyhu533 and others added 5 commits July 30, 2020 22:20

Update test_forward.py

bbb9c33

Update test_forward.py

7fee8c0

Update test_forward.py

77a6553

Merge branch 'relay_mx_np' of https://github.com/sandyhu533/incubator…

89f6937

…-tvm into relay_mx_np

Update test_forward.py

45e3e55

yzhliu approved these changes Aug 21, 2020

View reviewed changes

yzhliu merged commit 4c728d5 into apache:master Aug 21, 2020

ZihengJiang mentioned this pull request Sep 25, 2020

TVM v0.7 Release Note Candidate #6486

Closed

sxjscience mentioned this pull request Oct 20, 2020

[FRONTEND][MXNET 2.0] Support three missing operators: _npi_subtract_scalar, _npi_stack, _npi_advanced_indexing_multiple #6721

Closed

sxjscience mentioned this pull request Dec 31, 2020

[Frontend][MXNet] Importer Missing Operators #7186

Closed

3 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[RELAY][MXNET][FRONTEND] add support for MXNET numpy operators #6054

[RELAY][MXNET][FRONTEND] add support for MXNET numpy operators #6054

sandyhu533 commented Jul 14, 2020 •

edited

Loading

sxjscience Jul 15, 2020

yzhliu Jul 17, 2020

yzhliu Jul 17, 2020

yzhliu Jul 17, 2020

yzhliu Jul 17, 2020

sandyhu533 Jul 20, 2020

szha Jul 20, 2020

leezu Jul 20, 2020

yzhliu Jul 21, 2020

tqchen Jul 21, 2020

tqchen commented Jul 21, 2020

yzhliu Jul 28, 2020

yzhliu Jul 28, 2020

yzhliu Jul 28, 2020

yzhliu commented Jul 30, 2020

yzhliu commented Aug 21, 2020

sxjscience commented Aug 21, 2020

		@@ -1350,7 +1351,7 @@ def verify(batch, seq_length, num_heads, head_dim):
		verify(1, 10, 4, 16)
		verify(3, 10, 6, 8)

[RELAY][MXNET][FRONTEND] add support for MXNET numpy operators #6054

[RELAY][MXNET][FRONTEND] add support for MXNET numpy operators #6054

Conversation

sandyhu533 commented Jul 14, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tqchen commented Jul 21, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

yzhliu commented Jul 30, 2020

yzhliu commented Aug 21, 2020

sxjscience commented Aug 21, 2020

sandyhu533 commented Jul 14, 2020 •

edited

Loading