v4.1.2

mkturkcan/DARTv4.1.2Apr 6, 2026by mudler

AI Summary

A patch release that improves chat retry logic and introduces speculative decoding configuration for the llama.cpp backend.

Key Highlights

Added speculative decoding settings
Fixed autoparser logprobs passing
Fixed chat retry logic to handle chatdeltas and tooldeltas correctly

New Features

Speculative decoding settings

Full Release Notes

<!-- Release notes generated using configuration in .github/release.yml at master -->

## What's Changed
### Bug fixes :bug:
* fix(autoparser): correctly pass by logprobs by @mudler in https://github.com/mudler/LocalAI/pull/9239
* fix(chat): do not retry if we had chatdeltas or tooldeltas from backend by @mudler in https://github.com/mudler/LocalAI/pull/9244
### Exciting New Features 🎉
* feat(llama.cpp): wire speculative decoding settings by @mudler in https://github.com/mudler/LocalAI/pull/9238
### Other Changes
* Update index.yaml and add Qwen3.5 model files by @ER-EPR in https://github.com/mudler/LocalAI/pull/9237
* chore: :arrow_up: Update ggml-org/llama.cpp to `761797ffdf2ce3f118e82c663b1ad7d935fbd656` by @localai-bot in https://github.com/mudler/LocalAI/pull/9243
* chore: :arrow_up: Update leejet/stable-diffusion.cpp to `7397ddaa86f4e8837d5261724678cde0f36d4d89` by @localai-bot in https://github.com/mudler/LocalAI/pull/9242
* docs: :arrow_up: update docs version mudler/LocalAI by @localai-bot in https://github.com/mudler/LocalAI/pull/9241


**Full Changelog**: https://github.com/mudler/LocalAI/compare/v4.1.1...v4.1.2