strisunshinewentingwang/pytorch

Pull Requests Commits

[functorch] Added pointwise backward batching rules (elu_backward, hardshrink_backward, logit_backward, log_sigmoid_backward, gelu_backward, softplus_backward)

Chillee committed 2 years ago

fddaff44

[functorch] Added median/nanmedian batching rules

Chillee committed 2 years ago

70d3b160

[functorch] Added some fft batching rules

Chillee committed 2 years ago

d02e227c

[functorch] Added any.dim, count_nonzero.dim_IntList, cummax, cummin, linalg_vector_norm, logcumsumexp, nansum, nansum.dim_IntList, and mvlgamma

Chillee committed 2 years ago

d741c68d

[functorch] fix some errors in reduction batching rules

Chillee committed 2 years ago

854d5380

[functorch] Switched over to boxed implementations for reductions :)))))))

Chillee committed 2 years ago

11bd7e33

[functorch] whups

Chillee committed 2 years ago

2fb4a597

[functorch] Added boxed pointwise batching rule and almost finished pointwise (addcdiv, addcmul, clamp.Tensor, frexp.Tensor, lerp.Scalar, lerp.Tensor, log_sigmoid_forward, polygamma)

Chillee committed 2 years ago

6a7a022f

[functorch] Specialization cache (pytorch/functorch#99)

zou3519 committed 2 years ago

f9144e94

[functorch] Run python key tests in CI (pytorch/functorch#98)

zou3519 committed 2 years ago

38803ae0

[functorch] removed OP_DECOMPOSE

Chillee committed 2 years ago

a0a33697

[functorch] Started decomposing all ops that showed up in tests and also didn't throw errors

Chillee committed 2 years ago

4dd2f09f

[functorch] Added decomposition stops for all of the composite ops we haven't implemented yet (won't do anything until we decompose by default)

Chillee committed 2 years ago

e736a683

[functorch] fix embarassing bug causing builds to fail (I blame Copilot)

Chillee committed 2 years ago

3dd4196d

[functorch] Added all of the upsample no overload functions too (not sure if these ever even show up...)

Chillee committed 2 years ago

0884b8a5

[functorch] Fixed pythonkey tests

Chillee committed 2 years ago

eb26b8b9

[functorch] Updated some op annotations

Chillee committed 2 years ago

b178cf68

[functorch] fix CI hopefully

Chillee committed 2 years ago

f91c37db

[functorch] Finished rest of easy pointwise operators (16 of them)- bitwise_and, heaviside, hypot, gcd, igamma, igammac, lcm, nextafter, polar, rrelu_with_noise, bitwise_not, glu, logit, mish, threshold, copysign (kinda)

Chillee committed 2 years ago

b04e50d4

[functorch] updated the op DB

Chillee committed 2 years ago

6136d66b

[functorch] Added operator analysis generator

Chillee committed 2 years ago

eae7765b

[functorch] Remove the nondeterministic tests from xfail...

zou3519 committed 2 years ago

ac22e461

[functorch] Use new xfail mechanism in test_vmap_exhaustive; turn on CUDA for that test

zou3519 committed 2 years ago

4afc9a53

[functorch] New expected failure mechanism for test_ops.py

zou3519 committed 2 years ago

4d91cead

[functorch] Fix matmul batch rule; use decomposition from core instead of our own

zou3519 committed 2 years ago

82a7a6a8

[functorch] Fix softmax_backward batch rule

zou3519 committed 2 years ago

49b157f0

[functorch] Fix CI

zou3519 committed 2 years ago

fa50f1e2

[functorch] Adds one_hot decomposition, scatter batch rule (pytorch/functorch#93)

zou3519 committed 2 years ago

da8b20fc

[functorch] Improve dynamic op error message

zou3519 committed 2 years ago

f7b8cc64

[functorch] Add newline to eof

zou3519 committed 2 years ago

1a44e998

Newer Older