lemonade-sdk/lemonade v11.9.0 on GitHub

Headline

This quality release fixes two top bugs: allowed_origins no longer blocks remote use, and the fixed 120s prefill timeout is now a 600s default that is configurable via global_timeout.
A new experimental llamacpp-hrx backend runs AMD's HRX runtime in early access on AMD gfx1100/gfx1151 GPUs on Linux.
The pinned llama.cpp backend version was updated to add Qwen3-Next support.
Models placed in chat/, embeddings/, and reranking/ subfolders of extra_models_dir are now classified by type and listed individually.

Breaking Changes

extra_models_dir models in reserved folders (chat/, embeddings/, reranking/) are now listed individually per file instead of as a single extra.<directory> model; legacy extra.chat/extra.embeddings/extra.reranking IDs still resolve as hidden aliases to the first file alphabetically.
The LEMONADE_ALLOWED_ORIGINS environment variable is deprecated in favor of the allowed_origins config/CLI option and auto-migrates at startup with a warning; migrate to the new option, as the environment variable is only supported temporarily.

Lemonade Server

Operating System	Downloads
Windows	lemonade.msi
Ubuntu 24.04+	Launchpad PPA
Debian 13 (x86_64)	lemonade-server_11.9.0-debian13_amd64.deb
Debian 13 (ARM64)	lemonade-server_11.9.0-debian13_arm64.deb
Fedora 43 (x86_64)	lemonade-server-11.9.0-fc43.x86_64.rpm
Fedora 43 (ARM64)	lemonade-server-11.9.0-fc43.aarch64.rpm
Fedora 44 (x86_64)	lemonade-server-11.9.0-fc44.x86_64.rpm
Fedora 44 (ARM64)	lemonade-server-11.9.0-fc44.aarch64.rpm
macOS	Lemonade-11.9.0-Darwin.pkg

Other platforms? See our Installation Options for Docker, Snap, Arch, Debian, and more.

Embeddable Lemonade

Portable binaries for bundling into your own installer. Run lemond ./ as a subprocess.

Platform	Download
Ubuntu x64	lemonade-embeddable-11.9.0-ubuntu-x64.tar.gz
Ubuntu arm64	lemonade-embeddable-11.9.0-ubuntu-arm64.tar.gz
Windows x64	lemonade-embeddable-11.9.0-windows-x64.zip
macOS arm64	lemonade-embeddable-11.9.0-macos-arm64.tar.gz

What's Changed

Thanks @AaronStGeorge, @Bekhouche, @GabrielReusRodriguez, @Geramy, @abn, @anditherobot, @bitgamma, @fl0rianr, @github-actions, @jeremyfowers, @kenvandine, @superm1, @vaibhav8a, @zaneni6 for your awesome contributions to this release!

Click to expand changelog

Update FLM version to 1.0.3 by @zaneni6 in #3381
Add checkpoints to bench output by @bitgamma in #3359
feat(logging): add log rotation and transition default server posture to stdout/stderr by @abn in #2864
server: fix MTP draft companion handling by @fl0rianr in #3368
docs: fix typo maintainance -> maintenance by @vaibhav8a in #3383
fix: recover cache and models lost by 11.8.0 FHS split (#3387) by @superm1 in #3393
ci(sdcpp): run the CPU validation leg only on schedule/workflow_dispatch by @jeremyfowers in #3406
test(gpu-hang): scale request timeouts to the model-operation budget by @jeremyfowers in #3366
feat(backends): hrx backend [experimental] by @AaronStGeorge in #3374
fix(routing): llm classifier no longer treats has_tools as risk by @Bekhouche in #3330
fix(test): pre-download the model in test_gpu_hang_retry before the timed load by @jeremyfowers in #3408
fix(server): classify extra models using reserved directories: chat, embeddings, reranking instead of treating all model as chat by @anditherobot in #3333
fix(server): join the directory watcher before ModelManager members die by @Geramy in #3453
Update llama.cpp to b10723 by @github-actions[bot] in #3452
Fix(#3336) improve validate llamacpp.py by @GabrielReusRodriguez in #3379
fix(ci): resolve the newest llama.cpp nightly tag instead of /releases/latest by @jeremyfowers in #3450
ci(sdcpp): raise lemond global_timeout to 3600s for validation legs by @jeremyfowers in #3404
fix(server): log unauthorized origin rejections by @abn in #3418
ci(distro): retry transient mirror and HF fetch failures by @jeremyfowers in #3365
chore: Version bump to 11.9 by @kenvandine in #3454
fix(server): allow same-origin HTTP requests matching request Host by @abn in #3419
feat(server): add allowed_origins to global configuration schema by @abn in #3420

New Contributors

@vaibhav8a made their first contribution in #3383
@AaronStGeorge made their first contribution in #3374

Full Changelog: v11.8.1...v11.9.0

Windows installers are signed. Free code signing provided by SignPath.io, certificate by SignPath Foundation. See our Code Signing Policy.