Repository navigation
Feat/scout adoptions - #21
Merged
Merged
Conversation
Upstream head: b3c27359975ea4fb0f400785de3fc2729a35708a CUDA memory and restore validation pending.
Upstream head: f69225ed3e9221f014997365b652cbddc76cba4f CUDA build and target GPU A/B pending.
Upstream head: 6548710 Preserve the local embedded MTP trim. Target model validation pending.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Reduce prompt-checkpoint memory by omitting draft KV, recapture CUDA graphs after stable replay, and update d2t loading guards while preserving the embedded MTP trim.
Ports ggml-org#29509, ggml-org#29768 and updates ggml-org#29143. Adds long-context recall and parallel MTP acceptance probes, validation instructions and scout records. Existing GDN and fused gate/up carries remain unchanged. CUDA validation is pending.