Home

llama.cpp Upstream Support Status

Welcome to the llama.cpp upstream support status page for SpacemiT SOCs!

This page summarizes the current llama.cpp support status, validation progress, and upstream patch work on SpacemiT platforms.

Overview

llama.cpp is a lightweight inference framework for large language models. This section collects the main documentation, development branches, and platform information referenced throughout this page.

• Doc:
  • build-riscv64-spacemit
  • spacemit-mtmd-plugin
• Rolling patches file:
  • ggml-cpu/spacemit
  • spacemit-mtmd-plugin
• Support SOCs
  • SpacemiT-K1
  • SpacemiT-K3

Module Support Status

GGML Quantization Support For Matrix

Quantization Type	X60/X100	A100
Q2_K		✔️
Q3_K		✔️
Q4_0	✔️	✔️
Q4_1	✔️	✔️
Q4_K	✔️	✔️
Q5_0		✔️
Q5_1		✔️
Q5_K		✔️
Q6_K		✔️
Q8_0		✔️

Upstrem Component Status

All these patches will be upstreamed incrementally.

Platform	Component	Submitted time	Status	Link	Owner	Comments
K1	Add SpacemiT backend	Sep 29, 2025	Merged	15288	alex-spacemit
K3	Refactor SpacemiT backend and add K3 support	May 14, 2026	Merged	22863	alex-spacemit
K3	Optimize FlashAttention with IME2		WIP		co-seven

Local Component Status

These components are maintained as local extensions and will continue to be updated as needed.

Component	Submitted time	Status	Link	Owner	Comments
Extend compatibility to a broader range of multimodal vision models		WIP		co-seven

Monthly Update Log

Month	Summary	Updated by
2026-05	Initial wiki template created	alex-spacemit yutingnie