Repository navigation
fix: correct Bailing MoE Linear prefill state contract - #2299
Merged
Merged
Conversation
Disable chunked prefill because no-ID multi-token forwards reset model-owned recurrent and KV state while preserving stable-ID continuation. Add a loaded six-layer regression for independent fallback reset and per-sequence continuation. The source passed independent review; focused CUDA validation is running under the orchestrator. Refs #2287
inureyes
added
status:review
Preserve the newly merged MLX pin and overlay baseline declarations before validating the model prefill correction against current main. The model-specific implementation and synthetic regression are unchanged.
Merge current main through the mlxcelverse migration and revalidate loaded recurrent and KV state on MLX 243bdf09. All 36 focused CUDA tests pass, including the six-layer request-state contract. CUDA Clippy with warnings denied, formatting, and diff checks pass. Refs #2287
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Disable cache-level chunked prefill for Bailing MoE Linear because multi-token no-ID forwards intentionally reset model-owned recurrent and KV state. Add a loaded six-layer regression proving independent prompts reset every recurrent layer while stable sequence-ID chunks retain state.
Integrate main
b159c89d, including MLX243bdf09and mlxcelverse, without source conflicts. The rebuilt CUDA test-fast binary passes all 36 Bailing MoE Linear tests. CUDA Clippy with warnings denied, formatting, and diff checks pass. Earlier shared integration checks passed 123 Python tests and 49 WebUI fixtures; bilingual reports preserve their provenance and record the fresh final-pin run.Review found no remaining blocking implementation, security, or performance issue. DirectEngine supplies a stable sequence ID, but the capability is global: returning false makes every honoring caller use a whole-prompt graph, which can require more transient memory for long prompts. Tests load synthetic weights through the model loader; no external-checkpoint inference claim is made.
Closes #2287