Why Does llama.cpp’s Own API Give Xiaomi’s MiMo V2.6 Flash 9x the KV Cache?
Last Updated on September 25, 2026 by Editorial Team
Author(s): Chew Loong Nian – AI ENGINEER
Originally published on Towards AI.
At 128K tokens the model needs 2.96 GiB of cache. libllama’s default context settings would allocate 27.19 GiB, llama-cpp-python’s 28.31. Two flags fix it.
If you load Xiaomi’s new MiMo V2.6 Flash through llama.cpp’s C API, you get a KV cache 9.2x larger than the model needs at 128K context; through llama-cpp-python, 9.6x.
The article explains that although MiMo V2.6 Flash is designed for long context with only 9 of its 48 layers using global attention (the rest using a 128-token sliding window), the cache becomes massively oversized when code relies on llama.cpp defaults. It details how llama.cpp’s own cache-sizing rules interact with Xiaomi’s layer configuration, shows how a script can calculate the expected KV cache sizes, and verifies the results against measured llama.cpp logs. The root cause is traced to the `swa_full` default: the llama.cpp C API and llama-cpp-python inherit a full-size SWA cache (and in Python also defaults flash-attention settings differently), while the CLI tools use a smaller cache unless `–swa-full` is provided. Finally, it provides the practical fixes—set `swa_full=False` and ensure flash attention is enabled—and summarizes the trade-off: the full-size cache enables more advanced prompt caching, while the small cache dramatically reduces memory use (e.g., 128K context costing 2.96 GiB instead of 27+ GiB).
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Related posts
Recent Posts
