strisunshinewentingwang/pytorch

Pull Requests Commits

[MPS] Fix LSTM backward and forward pass (#95137)

alexdremov committed 1 year ago

b9e95158

[MPS] Fix view op slicing for 2nd dim in case of 0 offset (#95381)

DenisVieriu97 committed 1 year ago

86efa104

Update docs that Parameters are immune to no_grad mode (#95232)

janeyx99 committed 1 year ago

5783cee2

Add knobs for globally turning off 0/1 specialization and duck shaping (#95352)

ezyang committed 1 year ago

af202aea

Stop subclassing sympy Symbol (#95313)

ezyang committed 1 year ago

94fd063f

Add warn-once deprecation warning to legacy sparse constructors (#94850)

pearu committed 1 year ago

cece63f1

[autograd] disable backward/grad for complex scalar output (#92753)

kshitij12345 committed 1 year ago

3b966a6c

[Dynamo] No graph break on calling dict & collections.OrderedDict() (#95250)

yanboliang committed 1 year ago

b5ff41a4

std/var: support floating point correction value (#94073)

peterbell10 committed 1 year ago

bc438af6

SymFloat: Expose comparison operators in C++ API (#94812)

peterbell10 committed 1 year ago

56aed2a6

using float type to do the computation of norm reduce for cpu half and bfloat16 dtype (#95166)

XiaobingSuper committed 1 year ago

5730cabd

[DCP] Update DCP to use the updated FSDP optim state_dict APIs (#95303)

wz337 committed 1 year ago

6912cf40

Fix OOMing periodic shards (#95246)

ZainRizvi committed 1 year ago

c97275ac

[PTD][DCP] Add fsdp checkpoint example (#95258)

wz337 committed 1 year ago

bdb78e52

[vision hash update] update the pinned vision hash (#95340)

pytorchmergebot committed 1 year ago

c594a32f

[SDPA] Fix bug in parsing scaled_dot_product_attention arguments (#95311)

drisspg committed 1 year ago

29c235e5

use 4 warps for small block config in mm (#95339)

ngimel committed 1 year ago

8e391c73

[inductor] enable `test_nll_loss_forward_dynamic_shapes` (#95176)

nkaretnikov committed 1 year ago

ba8ff4be

Fix disbale typos (#95322)

kit1980 committed 1 year ago

f98733e9

Bugfix nested mem_efficient path in SDPA when E_qk != E_v (#95330)

mikaylagawarecki committed 1 year ago

586ac98c

[FSDP][Docs] Re-add why reg. post-bwd hook on 1st forward (#95326)

awgu committed 1 year ago

78175cee

Avoid FPE when running batch norm with zero batch size. (#95324)

ezyang committed 1 year ago

f247129f

coreml_delegate - Add input shape in error when throwing from predicting (#95249)

kylesyoon committed 1 year ago

a257486b

[pt2][inductor] switch dinfo representation (#95302)

nmacchioni committed 1 year ago

3ebab9ae

Preserve meta["val"] on export (#95314)

ezyang committed 1 year ago

ca7eb1ba

Second part of splitting #91254 in two (#92749)

Fabio Rocha committed 1 year ago

f6f413c6

[BE] Simplify `Source.is_nn_module`; add some types (#95292)

awgu committed 1 year ago

cbac56e2

Make fx.Transformer.get_attr call tracer to preserve node.meta (#95245)

ydwu4 committed 1 year ago

674ef1f9

Update isend/irecv warning messages for nccl (#95236)

H-Huang committed 1 year ago

c0fa0669

Implement sparse semantics support in gradcheck (#94714)

pearu committed 1 year ago

7ac511c2

Newer Older