docs(install): document NCCL dev libraries for multi-GPU tensor parallelism - #90
Open
alogotron wants to merge 1 commit into
Open
docs(install): document NCCL dev libraries for multi-GPU tensor parallelism#90alogotron wants to merge 1 commit into
alogotron wants to merge 1 commit into
Conversation
calvarado2004
added a commit
to calvarado2004/FreeToken
that referenced
this pull request
Aug 23, 2026
Implement the deployment feedback from the TP community testing: discover either an unversioned NCCL development link or a versioned runtime bundled with PyTorch, and fail with the searched paths before spawning ranks when neither exists. Expose a dedicated distributed/rendezvous port while preserving the existing API-port-plus-one default. Wire the DeepSeek-V4 test harness to the explicit port it already checks, so its collision guard and the engine use the same endpoint. Follow-up-to: FlashML-org#70 Builds-on: FlashML-org#90
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Adds the NCCL development libraries as a documented prerequisite for
--tensor-parallel-size > 1.On a clean install (Ubuntu 24.04, PyPI wheels, CUDA 13 toolkit via NVIDIA apt repo), the first TP>1 launch fails during the JIT link step:
Resolved with:
install.mdpreviously listed only driver r580+ / CUDA 13 and thenvccrequirement. Surfaced while testing TP=2 on 2x RTX 3090; also reported as a secondary finding in #29.