Skip to content

recipe: llama-cpp-python 0.3.37 (LFM3 support) - #136

Merged
ndonkoHenri merged 1 commit into
flet-dev:mainfrom
ndonkoHenri:llama-cpp-python-0.3.37
Nov 16, 2026
Merged

ndonkoHenri merged 1 commit into
flet-dev:mainfrom
ndonkoHenri:llama-cpp-python-0.3.37

Conversation

@ndonkoHenri

Copy link
Copy Markdown

Bumps llama-cpp-python 0.3.36 → 0.3.37. This release bundles llama.cpp a956a852c, which includes the lfm3 architecture added in b11960, so LiquidAI's LFM3 GGUF models now load. Requested in flet-dev/flet#7033.

Changes

  • meta.yaml: version 0.3.37. mobile.patch applies unchanged, and upstream's bindings already match the new llama.h structs.

Validation

  • CI (run 43776942105): Android + iOS, Python 3.12/3.13/3.14, all on-device tests pass.
  • A Flet 1.1.2 app on these wheels loaded LFM3-1.7B-Q4_K_M.gguf (1.1 GB) and generated a correct reply on an iOS simulator and an Android arm64 emulator (architecture reported as lfm3).

Consumer notes

Apps can now run LiquidAI LFM3 models on device with llama-cpp-python==0.3.37. The smallest LFM3 GGUFs are about 1.1 GB, so download them into app storage on first run rather than bundling them with the app.

@ndonkoHenri
ndonkoHenri merged commit 126f8c8 into flet-dev:main Nov 16, 2026
12 checks passed
@ndonkoHenri
ndonkoHenri deleted the llama-cpp-python-0.3.37 branch November 16, 2026 09:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant