Add xccl transport for device-resident put/get on Intel XPU#170
Open
songhappy wants to merge 1 commit into
Open
Add xccl transport for device-resident put/get on Intel XPU#170songhappy wants to merge 1 commit into
songhappy wants to merge 1 commit into
Conversation
XPU host: selector picks the new XcclTransportBuffer above Gloo; tensors stay on xpu end-to-end. CUDA hosts unchanged (xccl_available() returns False). Also gates shared_memory + torchcomms/cache for XPU, generalizes example/torchstore_rl.py to autodetect device, and adds example/torchstore_state_dict.py — multi-tensor state_dict round-trip mirroring GRPO weight-sync.
meta-cla
Bot
added
the
CLA Signed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
XcclTransportBufferkeepsts.put/ts.gettensors onxpuend-to-end. Selector picks it above Gloo on XPU hosts; CUDAbehavior is unchanged (
xccl_available()returns False).example/torchstore_rl.pyto autodetect cuda/xpu/cpu.example/torchstore_state_dict.py: trainer + generatoractors round-trip a multi-tensor LoRA-shaped state_dict — mirrors
RL weight-sync.
shared_memory.pyandtorchcomms/cache.py.Why
For RL weight sync (~600 MB state_dict per step), gloo stages
through CPU twice + TCP. xccl keeps bytes on XPU over oneCCL —
hundreds of ms → tens of ms per refresh.
Test plan
pre-commit runclean.torchstore_rl,torchstore_spmd,torchstore_state_dict) rc=0 on a 12×PVC node, torch2.12.0+xpu, bit-exact checksums.