strisunshinewentingwang/pytorch

Pull Requests Commits

Update MultiheadAttention module support key/value with different number of features and allow static key/value (#21288)

Guanheng Zhang committed 5 years ago

bb0f299f

Output sequence probability with CTC beam search, optional multiple output sequences (#21927)

warut-vijit committed 5 years ago

d684112e

EraseNumberTypes cleans itself up (#22461)

smessmer committed 5 years ago

830c6590

Remove build variable NCCL_EXTERNAL (#22467)

xuhdev committed 5 years ago

a6441c00

Create C2 operator to replace values in embedding table (#22279)

Alyssa Wang committed 5 years ago

34f950c8

include conv_op_impl.h from conv_dnnlowp_op.cc (#22458)

jspark1105 committed 5 years ago

040a4bd9

Warn on conditions that can trigger cuBLAS sgemm bug (#22034)

Brennan Vincent committed 5 years ago

474dec4b

Remove unnecessary ROCm detection code in Python scripts. (#22464)

xuhdev committed 5 years ago

f5b3f9ec

Fix compiler warnings (#22162)

smessmer committed 5 years ago

e68dc899

infer shape until no more change (#22425)

ChunliF committed 5 years ago

53a52f57

Add USE_MKLDNN_CBLAS build option. (#19014)

wuhuikx committed 5 years ago

07ef85e3

Use concrete types on call sites for Dict/List (#22004)

smessmer committed 5 years ago

6d587130

Rename macros and build options NAMEDTENSOR_ENABLED to BUILD_NAMEDTENSOR (#22360)

xuhdev committed 5 years ago

693871de

Pass LRU hash output evicted_values to SparseLookup (#21389)

Alyssa Wang committed 5 years ago

bb07f2d0

use feenableexcept when glibc is available (#22241)

Haixin Liu committed 5 years ago

869ce894

Fix empty_like for quantized tensors. (#21978)

gchanan committed 5 years ago

e74b0fc8

Revert D16088193: Refactored math tests to iterate over all math ops

Karl Ostmo committed 5 years ago

7235532d

Revert D16088191: Added math.log2 and hypot

Karl Ostmo committed 5 years ago

a845d02c

Fix PoolWindow crash from thread_local (#22405)

dkotfis-oculus committed 5 years ago

2dd71b18

ONNX Export LayerNorm

lara-hdr committed 5 years ago

7ca7edc3

Implement AdamW optimizer (#21250)

mjacar committed 5 years ago

a4b2f3e2

Numerical stability of embedding kernels (#22401)

madsbk committed 5 years ago

c9a84133

Added math.log2 and hypot

Chillee committed 5 years ago

b7687772

Refactored math tests to iterate over all math ops

Chillee committed 5 years ago

3d3d07b7

Make Gloo an optional c10d dependency (#22257)

pietern committed 5 years ago

0ffda97a

Remove the USE_MIOPEN build option as MIOpen is always used when built with ROCm. (#22420)

xuhdev committed 5 years ago

b9ede660

Remove hacky stub for quantized ops (#22388)

Dmytro Dzhulgakov committed 5 years ago

6721e67c

Fix the GPU trainer for NoneCalibration and RNN

xianjiec committed 5 years ago

2dd13233

Fix lint error in format_time() in throughput_benchmark.py and clean it up a bit. (#22424)

xuhdev committed 5 years ago

5bd97be3

Remove API-level guard on NeuralNetworks.h (#22429)

gkmhub committed 5 years ago

edd5b770

Newer Older