OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2026-01-15 12:15:51 +00:00
Code Issues Actions 53 Packages Projects Releases Wiki Activity
22,284 Commits 8,304 Branches 150 Tags
be757957bace28100e571ec7914765020be4a069
Commit Graph

830 Commits

Author SHA1 Message Date
Huan Gui
be757957ba Support softmax with D == 0 (#29167) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29167

As titled.

This fix is crucial as multi_channel splitting would create history that has no items (i.e., D == 0), which leads to flow failure.

Test Plan:
Unittest

flow test:

before fix: f148783160

after fix: f149082299

buck test mode/dev-nosan caffe2/caffe2/python/operator_test:softmax_ops_test

Reviewed By: xianjiec

Differential Revision: D18296081

fbshipit-source-id: e0bb2dc2c4e5b465e213f31e5c5ced3a7e1fd574
 2019-11-11 00:46:10 -08:00
Mike Ruberry
991c2ac383 Disables flaky test_rand_quantization (#29463) 
Summary:
See https://github.com/pytorch/pytorch/issues/28550.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29463

Differential Revision: D18405669

Pulled By: mruberry

fbshipit-source-id: 2984c3896a9260a06fbf052afb06e0cb8d28b53d
 2019-11-08 13:51:22 -08:00
Mike Ruberry
2f2a0d1607 Disables test_atomic_ops and testInputOrder (#29145) 
Summary:
These tests have been flaky for some time, see:

- https://github.com/pytorch/pytorch/issues/28179
- https://github.com/pytorch/pytorch/issues/9064

This PR disables them. The actual tests were added/updated 2+ years ago. It's unclear who, if anyone, would own them now.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29145

Differential Revision: D18327937

Pulled By: mruberry

fbshipit-source-id: d02731d662aff3545b581272e5ae8db4e3097d87
 2019-11-05 16:53:53 -08:00
Huan Gui
8a2dcff189 Add cuda version for operators BatchSparseToDense and BatchDenseToSparse (#29166) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29166

As titled

Test Plan:
unittest

 buck test  mode/dev-nosan  caffe2/caffe2/python/operator_test:batch_sparse_to_dense_op_test

Reviewed By: xianjiec

Differential Revision: D18197966

fbshipit-source-id: 7486300c509dd552ddb7484c2d83099f62878278
 2019-11-05 13:06:23 -08:00
Xinyi Zhang
5821b9bf0f Remove error logging of high empty range ratio 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28854

Reviewed By: xianjiec

Differential Revision: D18206695

fbshipit-source-id: 4ce471f0236b2ceaf54ba1b1ce96e193feca720b
 2019-10-30 12:55:25 -07:00
Huayu Li
793e2914e4 Support full id interations (#28769) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/28769

Support full id interaction.

Test Plan:
* unit-tests
  * buck test caffe2/caffe2/python/operator_test:pack_ops_test --
  * buck test caffe2/caffe2/fb/dper/layer_models/tests:sparse_nn_attention_test -- test_sparse_nn_full_id

* canary
  * apply SUM + full id with max_length as 20 on SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID: f147253340 (v1: f146340704)

# of embeddings for this features is 20:
{F219139816}

The corresponding ops: two lookups, which is as expected.
```
op {
  input: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_0/Repeat_0/sparse_lookup/w"
  input: "feature_preproc/output_features:SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM:values"
  input: "feature_preproc/output_features:SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM:lengths"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_0/Repeat_0/sparse_lookup/output"
  name: ""
  type: "SparseLengthsSum"
}
op {
  input: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/sparse_lookup/w"
  input: "feature_preproc/output_features:SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM:values"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/sparse_lookup/output"
  name: ""
  type: "Gather"
}
op {
  input: "feature_preproc/output_features:SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM:lengths"
  input: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/sparse_lookup/output"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/PackSegments/embedding_packed"
  name: ""
  type: "PackSegments"
  arg {
    name: "max_length"
    i: 20
  }
  arg {
    name: "pad_minf"
    i: 0
  }
}
op {
  input: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/PackSegments/embedding_packed"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/Reshape/reshaped_record"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/Reshape/old_shape"
  name: ""
  type: "Reshape"
  arg {
    name: "shape"
    ints: -1
    ints: 1280
  }
}
op {
  input: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/Reshape/reshaped_record"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_0"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_1"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_2"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_3"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_4"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_5"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_6"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_7"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_8"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_9"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_10"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_11"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_12"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_13"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_14"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_15"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_16"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_17"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_18"
  output: "nested/dot/SPARSE_AD_MEDIA_XRAY_V11_TOPIC_ID_AUTO_FIRST_X_AUTO_UNIGRAM/Pool_Option_1/Repeat_0/full_id/split/output_19"
  name: ""
  type: "Split"
  arg {
    name: "axis"
    i: 1
  }
}
```

Reviewed By: chonglinsun

Differential Revision: D18083520

fbshipit-source-id: f592fb7734dd4e3e712ba42dc0afcd0b32a4afa0
 2019-10-29 14:56:18 -07:00
Xinyi Zhang
f5ea2ca34a Reduce logging frequency for empty range tolarence 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28704

Reviewed By: xianjiec

Differential Revision: D18138828

fbshipit-source-id: 4f3c376502cb6e30b931217702c4ca537c9eb644
 2019-10-28 09:52:17 -07:00
Xinyi Zhang
2f16284231 change empty range tolorrance logging 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28489

Differential Revision: D18067322

fbshipit-source-id: 2096d1cce820f4ebe28db0045a2ddacc022e07da
 2019-10-23 09:39:39 -07:00
Xinyi Zhang
06bb74ce96 Tolerate small amount of embedding corruptions 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28371

Reviewed By: xianjiec

Differential Revision: D18031155

fbshipit-source-id: a51d2a62a919f032dc04372b30cf9071aa2dd629
 2019-10-21 16:23:25 -07:00
Jiang Wu
29f56eb920 Revert D17937850: Tolerate small amount of embedding corruptions 
Test Plan: revert-hammer

Differential Revision:
D17937850

Original commit changeset: e9c633768d98

fbshipit-source-id: 5c2c837c7867504392b19965d91a60cadd3b8101
 2019-10-19 14:17:01 -07:00
Xinyi Zhang
ca6ba06f95 Tolerate small amount of embedding corruptions 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/28299

Reviewed By: Wakeupbuddy

Differential Revision: D17937850

fbshipit-source-id: e9c633768d9819fd734ddd59017c33688ebbdcca
 2019-10-18 14:59:06 -07:00
Simran Suresh Motwani
d63d7ab997 Expose PiecewiseLinearTransform to PyTorch 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/26903

Test Plan: Unit Test

Reviewed By: bddppq

Differential Revision: D17585637

fbshipit-source-id: fe669aaf3301d7efb5c28ec0097945d55a71773d
 2019-09-27 12:49:04 -07:00
Jongsoo Park
8fb756d3b2 batch size 0 support in ChannelShuffle DNNLOWP op (#26858) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26858

Handle batch size = 0 in ChannelShuffle operator

Test Plan: CI

Reviewed By: jianyuh

Differential Revision: D17591041

fbshipit-source-id: 63373aa752406c1f38401c3e93d8e1954ce7281e
 2019-09-26 00:40:07 -07:00
Huan Gui
a8386d2a7d fix composite learning rate (#26227) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26227

In the previous implementation of composite lr, the lr_scale for each sub policy will be rewritten by the last lr_scale.

Due to another bug in unittest (where policy_lr_scale being the same for all sub policies), this bug was not detected by unittest...

Fix: add an additional field in CompositeLearningRateItem so that we store  lr_scale values for all sub policies

If fix unittest, the error in previous implementation:
https://fburl.com/testinfra/ikdbnmey

With the fix,
https://fburl.com/testinfra/m694ehl1

Test Plan:
unittest

buck test  caffe2/caffe2/python/operator_test:learning_rate_op_test -- test_composite_learning_rate_op

Reviewed By: chocjy, alex1o1o7cloud

Differential Revision: D17380363

fbshipit-source-id: 161e9cb71bb2ea7f0734a3361e270616057a08e4
 2019-09-18 17:34:17 -07:00
Qi Zhou
076eaf4ccf Exposing Fused8BitRowwiseQuantizedToFloat in PyTorch (#26080) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/26080

Will be used in c2 ctr_mbl_feed model to PyTorch conversion

Test Plan: Unit test

Reviewed By: yinghai

Differential Revision: D17337604

fbshipit-source-id: a90d9f5dc38301608d1562c6f2418e7f4616e753
 2019-09-12 12:36:33 -07:00
Frank Jiang
3be1745b3c Make SparseNormalize backwards compatible (#25660) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/25660

As title

Test Plan:
buck test caffe2/caffe2/python/operator_test:sparse_normalize_test
https://our.intern.facebook.com/intern/testinfra/testrun/5910974517813190

Reviewed By: boryiingsu

Differential Revision: D17187839

fbshipit-source-id: 1e5a6eaac0e825db4ae969540a1f689444070579
 2019-09-05 15:14:21 -07:00
Jongsoo Park
8199bb3dd3 add options to flush cache in SLS benchmarks (#25530) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/25530

Add an option to flush cache for more consistent benchmarking.

Test Plan:
buck run mode/opt caffe2/caffe2/fb/python/benchmarks:sparse_lengths_sum_4bit_benchmark -- --flush-cache
buck run mode/opt caffe2/caffe2/python/operator_test:sparse_lengths_sum_benchmark -- --flush-cache

Reviewed By: hyuen

Differential Revision: D17148087

fbshipit-source-id: 7eb782986676620254c1619a9a48c656cb1a6856
 2019-09-03 05:09:03 -07:00
Jongsoo Park
f1059d4e6a format sparse_lengths_sum_benchmark (#25529) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/25529

To prepare D17148087

Test Plan: Just formatting

Reviewed By: hyuen

Differential Revision: D17148085

fbshipit-source-id: faff90ee7dfec543d47037d20ce00f251144bc06
 2019-09-03 05:08:59 -07:00
Yanghan Wang
e34ef04301 register HeatmapMaxKeypoint with C10 (#25191) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/25191

registering as C10.

Test Plan: buck test mode/dev-nosan caffe2/caffe2/python/operator_test:heatmap_max_keypoint_op_test

Reviewed By: newstzpz

Differential Revision: D17056321

fbshipit-source-id: 989b72d7e3c9f23684b10d5fc9b98177ad4ee47b
 2019-08-27 20:13:57 -07:00
Frank Jiang
d7c6debc14 Remove gradient value as input from SparseNormalize op (#24357) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/24357

SparseNormalize does not need to know the gradient value to the lookup table, only the indices of the embeddings that need to be updated. By removing this input, we allow SparseNormalize to be used alongside SparseAdagradFusion

Differential Revision: D16809919

fbshipit-source-id: cc19692ba4dea8854663ae1ed8cf9365e90c99bc
 2019-08-19 14:47:09 -07:00
Yanghan Wang
3b22bbeb5b enable "keeps" from BoxWithNMSLimit and caffe2_fastrcnn_outputs_inference 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/24451

Reviewed By: newstzpz

Differential Revision: D16850259

fbshipit-source-id: 22f69d71a558d63c32a27d271a7557fc35a55176
 2019-08-19 10:54:22 -07:00
Yanghan Wang
ad64789a1e add aligned option to RoIAlign 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/23706

Reviewed By: ppwwyyxx

Differential Revision: D16615823

fbshipit-source-id: fd9152af8bc979cb04044413e66af349b032a99d
 2019-08-07 21:22:33 -07:00
Shali Jiang
15d3f0242b support Gather different indices for different examples in one batch (#23813) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/23813

Pull Request resolved: https://github.com/pytorch/pytorch/pull/23285

for example:

Inputs:
  data:
   [[[2 4 2 0],
     [0 1 2 0],
     [1 1 0 0]],
    [[3 4 1 3],
     [0 3 2 2],
     [4 1 0 4]]]

  idx:
    [[0 2],
     [0 1]]

outputs:
  [[[2 4 2 0],
    [1 1 0 0]],
   [[3 4 1 3],
    [0 3 2 2]]]

data and idx must have the same outer dimension

call Gather or BatchGather with argument match_outer=True

Reviewed By: huayuli00

Differential Revision: D16652485

fbshipit-source-id: 9e144e97a8d6fceaf3b5714df1534338068f4a10
 2019-08-07 21:14:30 -07:00
Michael Suo
a3c165f9d2 Revert D16452539: support Gather different indices for different examples in one batch 
Differential Revision:
D16452539

Original commit changeset: 7229489f4a9c

fbshipit-source-id: 010c177e551cb81521d2af84ce951bf964cdab44
 2019-08-05 10:22:01 -07:00
Shali Jiang
f87a4cc23f support Gather different indices for different examples in one batch (#23285) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/23285

for example:

Inputs:
  data:
   [[[2 4 2 0],
     [0 1 2 0],
     [1 1 0 0]],
    [[3 4 1 3],
     [0 3 2 2],
     [4 1 0 4]]]

  idx:
    [[0 2],
     [0 1]]

outputs:
  [[[2 4 2 0],
    [1 1 0 0]],
   [[3 4 1 3],
    [0 3 2 2]]]

data and idx must have the same outer dimension

call Gather or BatchGather with argument match_outer=True

Reviewed By: huayuli00

Differential Revision: D16452539

fbshipit-source-id: 7229489f4a9c02ee9f3c6a8a24bcd02925d96e07
 2019-08-04 21:17:49 -07:00
Jiexian Li
302adf1d20 add LambdaRank DCG Loss Option (#23679) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/23679
Full Canary: https://fburl.com/fblearner/sa1pkpya
Add LambdaRank DCG Loss Option
* when use_idcg_normalization == true, regular LambdaRank with NDCG loss
* when use_idcg_normalization == false, gradient and loss functions are not normalized by idcg.

Differential Revision: D16605459

fbshipit-source-id: a16f071e69516974e48d27bef4ca179019ca4ae7
 2019-08-02 11:47:46 -07:00
Jiexian Li
fc6aec9491 format only change (#23685) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/23685

format only changes.

Differential Revision: D16607482

fbshipit-source-id: 572afb59c6ff9f8a8842ba044fed6c87f8506843
 2019-08-02 11:47:42 -07:00
Levent Ertoz
6f01d13728 Implement dropout with replacement for id list features. (#22880) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22880

Implement sparse dropout with replacement value.

Reviewed By: xianjiec

Differential Revision: D16267012

fbshipit-source-id: 8c4878230f61bb3ac333291e2c6aaf2fbdc5f9ce
 2019-07-23 14:34:21 -07:00
Du Tran
d2ceab2766 update video input (#22471) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22471

update C2 video input with latest augmentation

Reviewed By: HengCV

Differential Revision: D16096127

fbshipit-source-id: bb07394e211cd52b50005d801b6d03250248ea9e
 2019-07-05 00:56:33 -07:00
Alyssa Wang
d9e15bccb0 Perform weight re-init for embedding table in sparse_lookup.py (#22348) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22348

This is the last step of LRU hash eviction weight re-init. This diff checks if there's evicted values in sparse_lookup, if so call op created in D15709866 to re-init the values for indicies in evicted_values. Also created gradient op for the operator. The gradient op just passes the output gradient as input gradient.

Reviewed By: itomatik

Differential Revision: D16044736

fbshipit-source-id: 9afb85209b0de1038c5153bcb7dfc5f52e0b2abb
 2019-07-03 10:33:40 -07:00
Duke Vijitbenjaronk
d684112ec9 Output sequence probability with CTC beam search, optional multiple output sequences (#21927) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21927

Add `OUTPUT_PROB` output to CTCBeamSearchDecoderOp to return a probability for each sequence.

Add argument to output top-k instead of top-1 decoded sequences.

Reviewed By: SuperIRabbit

Differential Revision: D15797371

fbshipit-source-id: 737ca5cc4f90a0bcc3660ac9f58519a175977b69
 2019-07-02 17:29:13 -07:00
Alyssa Wang
34f950c800 Create C2 operator to replace values in embedding table (#22279) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22279

This new operator is used for embedding table weight re-init. After we get the evicted indices, they will be the rows need reseting in embedding table. Then we can create a 1d tensor with default values, and apply this operator to copy the tensor to all evicted rows in embedding table

Will add gradient op in next diff

Reviewed By: itomatik

Differential Revision: D15709866

fbshipit-source-id: 2297b70a7326591524d0be09c73a588da245cc08
 2019-07-02 15:26:22 -07:00
Xiaomeng Yang
10e4137396 Optimize InstanceNormGradientOp (#22288) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22288

Optimize InstanceNormGradientOp

Benchmarks:

CPU with [N, C, H, W] = [128, 256, 56, 56],
NCHW order: 616ms -> 128ms
NHWC order: 1612ms -> 174ms

GPU with [N, C, H, W] = [128, 256, 112, 112],
NCHW order: 6450ms -> 37ms
NHWC order: 1419ms -> 82ms

Reviewed By: houseroad

Differential Revision: D16023630

fbshipit-source-id: 5af9bf1103cde2fc2bcb5cd5a057d039732f052e
 2019-07-01 15:10:17 -07:00
Xiaomeng Yang
29b53b0259 Fix bug in caffe2 transpose on GPU (#22233) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/22233

Fix bug in caffe2 transpose on GPU

Reviewed By: hl475

Differential Revision: D15994973

fbshipit-source-id: 542dc8757b51a6322fffa55826c1d4e32927398d
 2019-06-26 11:33:25 -07:00
Sungmann Cho
f59581218f Fix spelling errors (#21665) 
Summary:
alloctor -> allocator
excutable -> executable
excution -> execution
foward -> forward
initiaize -> initialize
paralell -> parallel
preprocesor -> preprocessor
tranpose -> transpose
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21665

Differential Revision: D15806155

Pulled By: soumith

fbshipit-source-id: d92b21ec8650a2b32f05faf9af0b7d2b073e992c
 2019-06-13 15:21:55 -07:00
David Zhang
696b2c89b4 Adding gradient to Boolean Mask operator (#21423) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21423

- add gradient for boolean mask
- add test for gradient checking

Reviewed By: BIT-silence

Differential Revision: D15640036

fbshipit-source-id: 79f40c6901e805bf1b8e9b01b57903e30b00f654
 2019-06-06 20:48:47 -07:00
David Zhang
cb2ec07fa2 ReshapeOp supports empty tensor (#21230) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21230

tsia; we support empty tensor with this diff for reshape operator

Reviewed By: jerryzh168

Differential Revision: D15583356

fbshipit-source-id: 6d44c04e95ca3546509bfb12102e29c878f9a7c7
 2019-06-06 15:02:11 -07:00
Hong Xu
da4f3629c5 Add missing shebangs to Python files with executable permissions. 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/21305

Differential Revision: D15613078

Pulled By: ezyang

fbshipit-source-id: 1fedf4368d65db406b617a51402ee8a20968aff7
 2019-06-06 10:53:40 -07:00
Yanghan Wang
81e70ffa19 fix bug of not using get_score_cls_index in BoxWithNMSLimitOp (#20868) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20868

When `input_boxes_include_bg_cls` is false (which means `input_scores_fg_cls_starting_id` is 0), It doesn't map the class index of score currectly when sorting and limiting the detections over all classes after nms.

Reviewed By: newstzpz

Differential Revision: D15472706

fbshipit-source-id: dc1e808b63ad09fb4bd95acf866771bb3fa92d69
 2019-05-24 22:31:01 -07:00
Yanghan Wang
371bd043d6 register ResizeNearestOp to C10 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/20928

Reviewed By: smessmer

Differential Revision: D15499661

fbshipit-source-id: 5af24d5c9d7ff739b8355e19dfe66b496bc026a5
 2019-05-24 14:39:11 -07:00
Kittipat Virochsiri
fd2aa93b37 Exposing LengthsSum/Mean/Max in pytorch (#20802) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20802

Need this for sequence model

Reviewed By: dzhulgakov

Differential Revision: D15448529

fbshipit-source-id: cd5abe3b689fc0e02feff10faf8cd61c99369f4f
 2019-05-22 13:55:19 -07:00
Huan Gui
fbdafdffa1 Move bucketize_op to open source 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/19952

Reviewed By: houseroad

Differential Revision: D15145552

fbshipit-source-id: e0074c878a5c164324a9cc477783285dedffd188
 2019-05-20 18:03:27 -07:00
Jongsoo Park
ea9c6e7581 eliminate FE_INVALID in unit test (#20502) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20502

Following D15307410 removing more floating point exceptions in unit tests

Reviewed By: hx89

Differential Revision: D15340930

fbshipit-source-id: 269fc75e0800bc9d39126767a0f3ca15cd8b0cad
 2019-05-16 21:55:28 -07:00
Yanghan Wang
373e6a78bf make box plus one a legacy argument in detection ops 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/20550

Reviewed By: newstzpz

Differential Revision: D15348610

fbshipit-source-id: 12b1e119e9bc9191ba9f2aa6d695ef215780c349
 2019-05-16 18:17:12 -07:00
Yanghan Wang
61012080c8 split and register CollectAndDistributeFpnRpnProposals with C10 
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/20509

Reviewed By: newstzpz

Differential Revision: D15302181

fbshipit-source-id: 7d3b29b667cd900f2976101f35200e1ee20b0f64
 2019-05-16 13:40:46 -07:00
Jongsoo Park
5f8e849d84 eliminate FE_INVALID in optimizer related operators and tests (#20501) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20501

Fixing unit tests related to optimizer related operators and tests

Reviewed By: hx89

Differential Revision: D15307410

fbshipit-source-id: e5400c26e08f26191ee542fe6b02e0a69bc4e1ae
 2019-05-16 08:23:46 -07:00
David Reiss
1891614aa5 Add GivenTensorInt16Fill (#20515) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20515

Needed by the upcoming quantized version of GenerateProposals

Reviewed By: dzhulgakov

Differential Revision: D14430952

fbshipit-source-id: ea852f04cc4b070f8fbe7a1e6535bba4d5b230fd
 2019-05-15 19:45:15 -07:00
Cheng Cheng
fd18b89c98 shape inference for learning rate op (#20020) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/20020

Add shape inference for LearningRate op. The output (lr) should have similar shape with input (iteration), but not the same type (float vs int).

Reviewed By: un-disclosed

Differential Revision: D15112300

fbshipit-source-id: 09969aefa15172a6f3c70cd9b2548e3020da5d7a
 2019-05-14 23:34:32 -07:00
Bilge Acun
3ee97183b0 ScaleBlobs Operator (#19660) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/19660

Implementation of aggregated Scale operator.
The operator takes a list of tensors as an input and scales all of them them with the argument float value.
The tensor sizes can be different, therefore bookkeeping of the sizes and pointers to the tensors are
necessary for the GPU version of the kernel.

Reviewed By: BIT-silence

Differential Revision: D14984233

fbshipit-source-id: 37cc97159a4f2c38cd6fff4f5710ab7d3a773611
 2019-05-08 17:57:33 -07:00
Jongsoo Park
42e9a619b3 add decay parameter in ref_adagrad (#15329) 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/15329

Add decay parameter to match with C++ Adagrad implementation.

Reviewed By: chocjy

Differential Revision: D13300991

fbshipit-source-id: db734df0202d8f5fd156f2742207d0b5a3aa7348
 2019-05-07 18:58:58 -07:00