microsoft/DeepSpeed

Pull Requests Commits

Add Japanese translation of Windows support blog (#6394)

tohtana committed 1 year ago

Verified 649b0785

fix fp16 Qwen2 series model to DeepSpeed-FastGen (#6028)

ZonePG committed 1 year ago

Verified e6fcc226

reduce cpu host overhead when using moe (#5578)

ranzhejiang committed 1 year ago

Verified 72608904

Long sequence parallelism (Ulysses) integration with HuggingFace (#5774)

samadejacobs committed 1 year ago

Verified 8b191d7c

Merge branch 'master' into tohtana/offload_zero_buffers

tohtana committed 1 year ago

Verified de2a8945

GDS Swapping Fix (#6386)

jomayeri committed 1 year ago

Verified b65ea506

improve offloading of lp params

Masahiro Tanaka committed 1 year ago

512e9c9f

Update linear.py compatible with torch 2.4.0 (#5811)

terry-for-github committed 1 year ago

Verified 96393f56

add verification in test

Masahiro Tanaka committed 1 year ago

93c5a906

fix offloading of lp grad

Masahiro Tanaka committed 1 year ago

37ffa021

add option for pin_memory and non blocking copy

Masahiro Tanaka committed 1 year ago

3f8179db

Bug Fix 5880 (#6378)

jomayeri committed 1 year ago

Verified 3831d4b5

Merge branch 'master' into tohtana/offload_zero_buffers

Masahiro Tanaka committed 1 year ago

837c06c8

DeepSpeed on Window blog (#6364)

tjruwase committed 1 year ago

Verified 01fe65b3

Pin transformers version on nv-nightly (#6002)

loadams committed 1 year ago

Verified ba0ab713

DeepNVMe GDS (#5852)

jomayeri committed 1 year ago

Verified 5f0d177f

loosen type hint

Masahiro Tanaka committed 1 year ago

05df37c5

reduce global reference to buffer

Masahiro Tanaka committed 1 year ago

126d9b7f

Merge branch 'master' into tohtana/offload_zero_buffers

tohtana committed 1 year ago

Verified 153a482f

Masahiro Tanaka committed 1 year ago

600c8226

Add and Remove ZeRO 3 Hooks (#5658)

jomayeri committed 1 year ago

Verified c2e3a706

add apis to offload states of model, optimizer, and engine

Masahiro Tanaka committed 1 year ago

58251044

Enable dynamic shapes for pipeline parallel engine inputs (#5481)

tohtana committed 1 year ago

Verified 1ab1928d

Move inf_or_nan_tracker to cpu for cpu offload (#5826)

BacharL committed 1 year ago

Verified 4d4ff0ed

add moe topk(k>2) gate support (#5881)

inkcherry committed 1 year ago

Verified 9a3ede70

move pynvml install to setup.py (#5840)

Rohan138 committed 1 year ago

Verified 30428d03

Update version.txt after 0.14.5 release (#5982)

loadams committed 1 year ago

Verified 3f6df9a2

Allow accelerator to instantiate the device (#5255)

nelyahu committed 1 year ago

Verified eb07d41f

Update xpu-max1100.yml with new config and add some tests (#5668)

Liangliang-Ma committed 1 year ago

Verified 4ba49dda

Z3: optimizations for grad norm calculation and gradient clipping (#5504)

nelyahu committed 1 year ago

Verified 6eed634e

Newer Older