deepspeedai/DeepSpeed

Pull Requests Commits

Merge branch 'master' into reyazda/fix-inference-api

RezaYazdaniAminabadi committed 4 years ago

Verified 79ee82f6

[zero2] zero_param_shapes: switch to round_robin_fp16_groups (#1240)

stas00 committed 4 years ago

Verified 7392e459

[see_memory_usage] fix deprecation (#1234)

stas00 committed 4 years ago

Verified 32e85eda

enable async io op on powerpc architectures (#1224)

adammoody committed 4 years ago

Verified 89b0fb42

[zero3] fix reference counting in backward over multiple forwards (#1227)

stas00 committed 4 years ago

Verified 3fa24208

[zero3] release tmp memory when consolidating fp16 weights take2 (#1223)

stas00 committed 4 years ago

Verified 81039463

Bump addressable from 2.7.0 to 2.8.0 in /docs (#1225)

dependabot[bot] committed 4 years ago

Verified 54bed327

ZeRO2-Offload: Disable copy overlapping (#1219)

tjruwase committed 4 years ago

Verified 97207c8c

formatting fix for release script

jeffra committed 4 years ago

d1a7a55e

bump to 0.4.4 + add release scripts

jeffra committed 4 years ago

dba0f431

revert part of #1220 (#1221)

stas00 committed 4 years ago

Verified bc451c09

[zero3] release tmp memory when consolidating fp16 weights (#1220)

stas00 committed 4 years ago

Verified 2660cc4d

[model weights] zero_to_fp32 multiple improvements (#1181)

stas00 committed 4 years ago

Verified 2a921069

enable cpu adam op on powerpc architectures (#1213)

adammoody committed 4 years ago

Verified f65ff908

json wrong format (#1210)

GeorgeSen committed 4 years ago

Verified 5652072e

improve debug (#1215)

stas00 committed 4 years ago

Verified 5127b2fa

[zero.Init] post_init partitining is to be run only by a child module (#1202)

stas00 committed 4 years ago

Verified 497b741f

[zero3] params_to_reduce isn't always there (#1214)

stas00 committed 4 years ago

Verified 91f58c06

Upgrade bundler to address CVE-2020-36327

jeffra committed 4 years ago

Verified aeea85b4

send input_mask to compute_attn func for checking

Reza Yazdani committed 4 years ago

7cb65cc9

Reza Yazdani committed 4 years ago

65f1fe82

Reza Yazdani committed 4 years ago

485201bf

use a dummy tensor if input_mask is none

Reza Yazdani committed 4 years ago

e628299f

fixing inference api for FP32 and non-masking GPT-based models

Reza Yazdani committed 4 years ago

db492f60

contiguous gradients should be set to True by default (#1199)

samyam committed 4 years ago

Verified c9fee821

jeffra committed 4 years ago

36126122

clean up logging (#1190)

stas00 committed 4 years ago

Verified a0292398

undo noise (#1191)

stas00 committed 4 years ago

Verified bc019a53

Fix bugs about non-contiguous tensor broadcasting (#1168)

hyunwoongko committed 4 years ago

Verified 429cbc89

introduce debug utils (#1136)

stas00 committed 4 years ago

Verified c0c4ebf1

Older