Commits

History for

Commits on Apr 12, 2023

[CPU support] Optionally bind each rank to different cores on host (#2881)

authored
0b5252b
feat: Add support for `NamedTuple` when sharding parameters [#3029] (#3037)

authored
29aea09

Commits on Apr 11, 2023

DeepSpeed Chat (#3186)

authored
47f9f13

Commits on Apr 10, 2023

deepspeed/runtime/utils.py: reset_peak_memory_stats when empty cache (#2803)

guoyejun
and
tjruwase
authored
6eca037

Commits on Apr 7, 2023

zero.Init() should pin params in GPU memory as requested (#2953)

tjruwase
and
mrwyattii
authored
4d27225

Commits on Apr 6, 2023

Fix for Diffusers 0.14.0 (#3142)
molly-smith
authored
c48c979

Commits on Apr 5, 2023

Recover shared parameters (#3033)

ShijieZZZZ
and
tjruwase
authored
30d9770
Update skip on torch in tests (#3136)
loadams
authored
ab1f32d
Update engine.py (#2826)

stas00
and
tjruwase
authored
bcb0353
Enable autoTP for bloom (#3035)

sywangyi
and
tjruwase
authored
6ba0024

Commits on Apr 4, 2023

Fix launch issue (#3137)
tjruwase
authored
fcb868e

Commits on Mar 31, 2023

Update DeepSpeed copyright license to Apache 2.0 (#3111)

mrwyattii
and
jeffra
authored
b361c72

Commits on Mar 30, 2023

Commits on Mar 28, 2023

Disable Stage 1&2 CPUAdam pathways (#3097)

mrwyattii
and
tjruwase
authored
4b6d7c1

Commits on Mar 27, 2023

update formatter version and style settings (#3098)
jeffra
authored
91d63e0

Commits on Mar 24, 2023

pre-commit check for torch.cuda in code (#2981)

delock
and
tjruwase
authored
090d49e
Empty ZeRO3 partition cache (#3060)

tjruwase
and
jeffra
authored
e80ae08

Commits on Mar 23, 2023

Goodbye Torch 1.8 (#3082)

mrwyattii
and
jeffra
authored
5cdf359
allow list (#3042)

satpalsr
and
jeffra
authored
5c2a81c
Fix nebula in save_16bit_model issue (#3023)

FreyaRao
and
CodeSkull-1
authored
a78d6b8

Commits on Mar 22, 2023

Remove bf16 from inference config dtye enum (#3010)

molly-smith
and
jeffra
authored
27e1b02

Commits on Mar 21, 2023

Assert mp_size is factor of model dimensions (#2891)

molly-smith
and
jeffra
authored
9ea0fdc

Commits on Mar 15, 2023

Improve loss overflow logs (#3008)

Quentin-Anthony
and
tjruwase
authored
ac2c9ff
Convert model parameters from generator to list. (#3017)

jomayeri
and
tjruwase
authored
94f7da2
[logger] implement warning_once (#3021)

stas00
and
tjruwase
authored
50a49e4
adding attribute checks for bf opt with zero (#3022)

jomayeri
and
tjruwase
authored
d7c925e

Commits on Mar 14, 2023

Update torch_checkpoint_engine.py (#3019)
stas00
authored
e355863

Commits on Mar 13, 2023

Fix buffer size for pipeline parallel and communication schedule (#2862)

authored
b528f50
ckpt: create directories in checkpoint_engine (#2988)

adammoody
and
tjruwase
authored
43d58d9

Commits on Mar 10, 2023

Fix Meta Tensor checkpoint load for OPT models (#2990)
lekurile
authored
3798e60

Commits on Mar 9, 2023

[zero] prevent poor configs from running w. zero-offload (#2971)
jeffra
authored
457850d

Commits on Mar 8, 2023

bug fix for skipping mbs (#2171)

rahilbathwal5
and
rajhans
authored
6379def

Commits on Mar 7, 2023