[XPU] Support XCCL on deepspeed side#7113

Closed

yisustc wants to merge 59 commits into

deepspeedai:masterfrom

yisustc:sy/xccl_enable

Commits on May 21, 2025

support XCCL on deepspeed side
yisustc
committed
Update gaudi2 nightly,ci to latest 1.20.0 build (#7093)

authored and
yisustc
committed
fix keep_module_on_host (#7112)

authored and
yisustc
committed
Add sequential pytest mark to TestNVMeCheckpointing to resolve pytest forked hangs (#7131)

loadams
authored and
yisustc
committed
Training multiple models (#7018)

authored and
yisustc
committed
Update CONTRIBUTING.md to reflect changes from CLA to DCO (#7135)

loadams
authored and
yisustc
committed
Avoid missing attr error (#7133)

tjruwase
authored and
yisustc
committed
Add conditional expression (#7119)

authored and
yisustc
committed
Unpin transformers version for most workflows (#7139)

loadams
authored and
yisustc
committed
Conditionally quote env vars (#7071)

authored and
yisustc
committed
Correct the BACKWARD_PREFETCH_SUBMIT mismatch (#7120)

authored and
yisustc
committed
Enhance Gaudi2 CI/Nightly Coverage with Model Parallelism and Linear Tests (#7146)

raza-sikander
authored and
yisustc
committed
Update container version that runs on A6000 tests. (#7153)

loadams
authored and
yisustc
committed
fix leak of z3 buffer

Masahiro Tanaka
authored and
yisustc
committed
hf tp+zero training doc. (#7151)

authored and
yisustc
committed
Avoid graph break by removing redundant requires_grad attr change (#7158)

authored and
yisustc
committed
Add destroy to tests to free memory (#7160)

tohtana
authored and
yisustc
committed
[NFC] Typo fix in SP layer. (#7152)

authored and
yisustc
committed
Link AutoTP blog in the front page (#7167)

authored and
yisustc
committed
fix `seq_parallel_communication_data_type` constant. (#7175)

stas00
authored and
yisustc
committed
Fix typos in GDS blog (#7177)

loadams
authored and
yisustc
committed
Variable batch size and LR scheduler (#7104)

authored and
yisustc
committed
Update version.txt after 0.16.5 release (#7180)

loadams
authored and
yisustc
committed
Cross layer overlapping for domino (#7178)

authored and
yisustc
committed
async tp allreduce (#7115)

authored and
yisustc
committed
Fix issue #5242 grad_norm and loss is nan (#7171)

authored and
yisustc
committed
Add qwen3 autotp support (#7187)

authored and
yisustc
committed
Update to new torch grad hook API: BF16Optimizer and Stage2 (#7189)

authored and
yisustc
committed
Reland perf fix for nan inf check (#7184)

authored and
yisustc
committed
Update to fix pydantic warning (#7193)

loadams
authored and
yisustc
committed
update dependencies version info (#7206)

inkcherry
authored and
yisustc
committed
HPU accelerator memory mapping is broken because of torch fill uninit memory (#7209)

oelayan7
authored and
yisustc
committed
Support complicated use cases with TiedLayerSpec (#7208)

authored and
yisustc
committed
Add defence for offload_states and reload_states w/o optimizer (#7211)

HollowMan6
authored and
yisustc
committed
DeepCompile for enhanced compiler integration (#7154)

authored and
yisustc
committed
Update version.txt after 0.16.6 release (#7218)

authored and
yisustc
committed
Fix release links (#7219)

authored and
yisustc
committed
Fix pass for z3 and profiler (#7222)

tohtana
authored and
yisustc
committed
Fix build on AMD GPUs (related to DeepCompile) (#7224)

HollowMan6
authored and
yisustc
committed
Add defence for DeepCompile w/o optimizer (#7225)

authored and
yisustc
committed
Pass `with_cuda` arg for jit_load in OpBuilder (#7226)

HollowMan6
authored and
yisustc
committed
Make sure it's not None before offloading contiguous_grad_buffer (#7227)

authored and
yisustc
committed
Update version.txt after 0.16.7 release (#7232)

loadams
authored and
yisustc
committed
Recommend using latest (#7233)

tohtana
authored and
yisustc
committed
[NFC] Fix comment related to SP group (#7234)

c8ef
authored and
yisustc
committed
Add cpu accelerator fp16 dtype support (#7207)

authored and
yisustc
committed
Update torch cpu test version

loadams
authored and
yisustc
committed
Revert "Update torch cpu test version"

loadams
authored and
yisustc
committed
Update CPU torch version to 2.7 (#7241)

loadams
authored and
yisustc
committed
Update README.md (#7246)

jizhang02
authored and
yisustc
committed
Fix compile error for nv_bloat162 (#7248)

authored and
yisustc
committed
add `Makefile` to ease maintenance (#7267)

stas00
authored and
yisustc
committed
Fix fp8 gemm (#7265)

authored and
yisustc
committed
[XPU] update xpu-max1100 CI workflow to torch 2.7 (#7284)

authored and
yisustc
committed
Fix issues XPU tests hit with extra-index-url (#7291)

loadams
authored and
yisustc
committed
Temporarily skip AIO tests due to an issue with runners (#7288)

authored and
yisustc
committed
rollback #6726 (#7258)

authored and
yisustc
committed
Update patch version after 0.16.8 release (#7296)

loadams
authored and
yisustc
committed
fix non-torch failure, if the torch version is too old
yisustc
committed