Skip to content

feat: add Laguna XS.2 & M.1 arch support (cherry-pick upstream #25165) - #62

Merged
turbo-tan merged 2 commits into
mainfrom
feat/laguna-arch
Jul 30, 2026
Merged

feat: add Laguna XS.2 & M.1 arch support (cherry-pick upstream #25165)#62
turbo-tan merged 2 commits into
mainfrom
feat/laguna-arch

Conversation

@turbo-tan

Copy link
Copy Markdown
Owner

Cherry-picks merged upstream PR ggml-org#25165 (Poolside Laguna XS.2 & M.1 arch). Clean apply, zero conflicts.

What's included

  • src/models/laguna.cpp — sigmoid-routed MoE, shared expert, QK-norm, YaRN/SWA RoPE
  • Arch registration in llama-arch.h/cpp, models.h
  • Chat templates for Laguna S/XS
  • Conversion script (conversion/laguna.py)
  • Test coverage in test-llama-archs.cpp, test-chat-auto-parser.cpp

Validated

  • Laguna-XS-2.1 BF16 → TQ3_4S (16 GB, 4.05 BPW)
  • Laguna-S-2.1 Q8_0 → TQ3_4S (56 GB, 4.02 BPW)
  • Benchloop (XS, 3090): overall 73.7, speed 96.4, coding 100.0, gen 196 tok/s
  • Hard86 (XS): 55/86 (64.0%), 202 tok/s
  • HE+ 0.762, MBPP+ 0.720 (XS)

@turbo-tan
turbo-tan merged commit 80e7a9b into main Jul 30, 2026
28 checks passed
@turbo-tan
turbo-tan deleted the feat/laguna-arch branch July 30, 2026 23:15
turbo-tan added a commit that referenced this pull request Aug 1, 2026
…rg#25165) (#62)

* feat: add Laguna XS.2 & M.1 arch support (cherry-pick upstream ggml-org#25165)

* ci: run check-requirements on GitHub-hosted runner

Assisted-by: opencode

---------

Co-authored-by: charpdev <charpdev@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants