Upgrade llama.rn to 0.13.0-rc.3 (llama.cpp b10588 -> b10829) - #901
Merged
Conversation
…829) Syncs the vendored llama.rn native dependency from 0.13.0-rc.1 to 0.13.0-rc.3 — a llama.cpp sync of b10588 -> b10829, plus an Android fix that probes HTP sessions for the real device count and keeps CPU out of device overrides so CPU paths can use optimized repacked buffers. New model support: hy-v4, qwen4exp, spark2-5, deepseek4v. The TypeScript surface of llama.rn is unchanged apart from src/version.ts, so no app code moves. Note: ios/Podfile.lock SPEC CHECKSUMS for llama-rn still reflects the rc.1 hash; it needs a pod install to recompute once the react-native-worklets-core resolution issue (pre-existing, unrelated to this bump) is addressed. Signed-off-by: dongjiang1989 <dongjiang1989@126.com>
Contributor
Author
|
cc @a-ghorbani PTAL, thanks |
Contributor
Author
|
Thanks @a-ghorbani |
Contributor
PR #901 benchmark results
poco-myron · cpu · 29 cells
poco-myron · hexagon · 29 cells
poco-myron · gpu · 29 cells
samsung-s23 · cpu · 29 cells
samsung-s23 · hexagon · 29 cells
samsung-s23 · gpu · 29 cells
poco-x7-klee · cpu · 29 cells
|
Owner
|
@dongjiang1989 I hope the numbers above are clear enough. We found a memory regression when Hexagon uses multiple sessions. see
I’ll update PocketPal to select a single HTP session to avoid it (issue #904). Hopefully after that we can merge this. |
Contributor
Author
Thanks @a-ghorbani When will the new release be published? |
a-ghorbani
approved these changes
Sep 10, 2026
a-ghorbani
left a comment
a-ghorbani
left a comment
Owner
There was a problem hiding this comment.
lgtm - llama.cpp changes introduced a HTP memory regression on devices with multiple sessions. #907 fixes this by selecting only a single HTP session.
Owner
i am cutting a release now, if everything passes, I guess we'll have it published in the stores in 1–2 days. |
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
2 participants
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.
Description
Upgrades the vendored
llama.rnnative dependency from0.13.0-rc.1to0.13.0-rc.3— a llama.cpp sync of b10588 → b10829.The TypeScript surface of
llama.rnis unchanged apart fromsrc/version.ts, so no app code moves.What's new in rc.3 (over rc.1)
hy-v4,qwen4exp,spark2-5,deepseek4vKnown follow-up
ios/Podfile.lockSPEC CHECKSUMS forllama-rnstill reflects the rc.1 hash; it needs apod installto recompute. The current localpod installfails on a pre-existing, unrelatedreact-native-worklets-coreresolution issue (reproduces onmainbefore this bump), so the checksum was left for a follow-up once that is addressed.Platform Affected
Checklist