Publish-only delta sync, re-derived on the disk-delta base#6
Closed
nanjiangwill wants to merge 9 commits into
Closed
Publish-only delta sync, re-derived on the disk-delta base#6nanjiangwill wants to merge 9 commits into
nanjiangwill wants to merge 9 commits into
Conversation
Ship only the changed bytes between weight syncs as a canonical HF delta checkpoint; rollout hosts apply it into a host-local checkpoint and reload via the vanilla update_weights_from_disk path. Replaces the NCCL delta transport from THUDM#1806 with a disk-only path that needs no engine-side delta support.
sync_local_checkpoint (was sync_weights) materializes the base lazily via the idempotent init_local_checkpoint instead of a background thread; record per-sync update time in update_weight_metrics; state the pre-read/pre-push hooks' purpose (non-POSIX filesystem coherence).
The actor's update_weights is already @timer-wrapped (perf/update_weights_time), so the per-sync total/publish/reload breakdown was duplicate instrumentation. Keep only the delta-specific metrics (density, wire bytes).
The delta scaffold reworked the update-weight args: delta requires --update-weight-transport=disk (was nccl-or-disk), needs --update-weight-local-checkpoint-dir, and the --update-weight-delta-dir compatibility alias is gone (the directory belongs to the transport, not the encoding). Drop the alias resolve/backfill/conflict tests, point the transport and colocate tests at the disk path, and cover the local-checkpoint requirement.
With the delta-dir alias gone, _resolve_update_weight_disk_dir no longer normalizes anything — it's a single transport-level check, so fold it into _validate_update_weight_args.
slime_validate_args validates everything else inline; the extracted _validate_update_weight_args was the lone exception. Fold it in and test it the same way as the other slime_validate_args checks (make_slime_validate_args).
Materialize the host-local checkpoint in a daemon thread at engine init so the one-time base copy overlaps sglang launch and the first rollout (which serves from init-loaded weights) instead of blocking the first delta reload. The first sync_local_checkpoint's init_local_checkpoint is idempotent and flock-guarded, so it either finds the copy done or blocks on the same lock — no join needed.
1ab3399 to
1499c3e
Compare
d70b68a to
f506c33
Compare
Add --rollout-endpoint-url to train against an elastic rollout fleet behind a single opaque HTTP endpoint (no per-engine handles). Generation routes to the URL; the disk-delta updater publishes each version to the shared disk dir and advances a `latest` pointer for the fleet to pull, instead of pushing via per-engine RPCs. Add-only: managed and external-addressed paths are unchanged. Requires delta mode + disk transport (the only cross-cluster channel) and is non-colocate. - abort: the endpoint cancels surplus when discarding; with --partial-rollout it drains so streaming tasks return their partials (closing each stream disconnects, which aborts the fleet-side request). Endpoint + --partial-rollout requires the streaming rollout, the only path that captures partials client-side. - streaming: record the weight version on an aborted partial so off-policy correction can weight it (update_from_meta_info is skipped without a finish_reason).
f506c33 to
94ac929
Compare
Author
|
Closing: jvmncs/rollout-endpoint is deprecated. The publish-only work now stacks on the disk-delta branch via #5 (disaggregated-rollout -> disk-delta-weight-sync). |
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
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.
Clean re-derivation of this branch's publish-only weight sync, rebuilt on top of the disk-level delta weight sync (PR to THUDM THUDM#2089) instead of the old
update_weight_from_distributed_delta.pythat THUDM#2089 removes.The large diff vs this branch is the base difference: it carries all of THUDM#2089's delta rewrite plus the disaggregated-rollout feature. The publish-only logic itself is ~90 lines (see #5) — on THUDM#2089's clean disk-delta, "publish" is just write the version + advance a
latestpointer, skip the engine RPCs.Dropped relative to this branch:
--update-weight-delta-publish-waitnext-sync/sync dual mode (publish is synchronous: write + announce, no in-flight drain).--update-weight-delta-root(unused).meta_info["weight_version"](captured intoSample.weight_versions), and routing/affinity belong to the fleet's router.Generation routing, placement, and abort follow the existing external-engine pattern (
ExternalRolloutServer). Intended to supersede the publish-only approach here once THUDM#2089 lands.