Commits pytorch/benchmark

Do to convert UnsupportedFakeTensorException into RuntimeError in runNode for proper graph breaking. (#120026)

laithsakka committed 2 years ago

49c4a5fb

Log more compilation time breakdown (#119865)

yanboliang committed 2 years ago

e194d216

Log the optimus graph transformation to the scuba (#2163)

mengluy0125 committed 2 years ago

f5b2e501

Remove the date constraint in the scheduled runs (#2167)

xuzhao9 committed 2 years ago

d7b17ab3

Propagate allow_non_graph_fake between get_fake_values_from_nodes and get_fake_values (#119731)

laithsakka committed 2 years ago

a4053bd3

Skip test_wrap_bad if run under pytest (#115070)

Flamefire committed 2 years ago

06ccf5ac

Deploy inductor_speedup as the servicelab benchmark

xuzhao9 committed 2 years ago

0907249c

Change the --no-optimize-ddp flag to reflect the latest usage (#119437)

fegin committed 2 years ago

dbc10979

Support FunctoolsPartialVariable::get_function, fix get_fake_values_from_nodes (#119435)

laithsakka committed 2 years ago

1ca89b11

Improve support for backwards hooks (#119525)

jansel committed 2 years ago

542c8765

Enable `--num-batch` by default (#2157)

xuzhao9 committed 2 years ago

62f4e9c6

Bisect new nightly error and fix the broken T4 workflow. (#2153)

xuzhao9 committed 2 years ago

18eafb4e

Further delay the A100 nightly run (#2158)

xuzhao9 committed 2 years ago

28275e21

Merge allow_in_graph/inline/skip trace rules check into trace_rule.lookup (#118971)

yanboliang committed 2 years ago

b003e293

Skip dynamo when inside a functorch context (#118901)

jansel committed 2 years ago

93c2136e

Make start compiling stack trace omit framework frames (#119251)

ezyang committed 2 years ago

46188bf4

Fix a bug in group_bench userbenchmark

xuzhao9 committed 2 years ago

71cc87aa

Add bisection interactive mode for debugging. (#2148)

xuzhao9 committed 2 years ago

ab07f168

Some minor type stub improvements (#118529)

ezyang committed 2 years ago

b30d690b

pytorch/benchmark 1/9