Repository navigation
unsloth: pin #243 to bound the glm5-next compute buffer - #244
Merged
Merged
Conversation
|
You have reached your Codex usage limits for security reviews. Please try again later. |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Pins #243 so the nightly prebuilts carry the glm5-next compute buffer fix. On GLM-5.3-Flash at
-c 244480 -np 4 --kv-unifiedthe CUDA compute buffer drops from 4939 MiB to 1153 MiB, which is what lets the model, its vision projector and the NextN MTP head fit a 96 GB card.pr-set.json: appendsunslothai#243 @ aba4a4c1d, after Carry what the dropped GLM-5-Next, Qwen MTP and readahead pins had beyond upstream #241.feature-checks.json: lists glm5-next : keep the indexer compute buffer bounded at long contexts #243 underunchecked. It changes memory layout only, with no observable output difference, and the size only shows on a CUDA host with a GLM-5-Next model.Testing
additive_merge.py, on b11368 and on b11435 (the newest tag): every pin merges, glm5-next : keep the indexer compute buffer bounded at long contexts #243 cleanly. Add TML Inkling architecture ggml-org/llama.cpp#25731 and kimi-k3 : the MoonViT-3d vision tower and full-size loading fixes #70 need the additive resolver, as before.-DLLAMA_FATAL_WARNINGS=ON.scripts/unsloth/test_*.pypass locally.