PulseAugur
EN
LIVE 20:13:08

llama.cpp v0.7.0 enhances GPU caching for faster MoE expert inference

By PulseAugur Editorial · [1 sources] ·

The latest release of llama.cpp, version 0.7.0, introduces a new GPU cache for Mixture-of-Experts (MoE) models. This feature, accessible via the --moe-cache-mib option, allows experts to run on the GPU from an LRU cache, potentially doubling decoding speed. Additionally, the expert cache can now be shared across multiple GPUs, enhancing performance for multi-GPU setups. AI

IMPACT Improves inference speed and efficiency for MoE models on consumer hardware.

RANK_REASON Release of a new version of an open-source inference library with performance improvements. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Mastodon — mastodon.social →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

llama.cpp v0.7.0 enhances GPU caching for faster MoE expert inference

How we ranked this

Signal score
12 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Release of a new version of an open-source inference library with performance improvements. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
infra, model release
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. Mastodon — mastodon.social TIER_1 English(EN) · whatsnew_fyi ·

    llama.cpp v0.7.0 New GPU cache for host-memory MoE experts via the --moe-cache-mib option, with experts run on the GPU from an LRU cache for up to 2.2x faster d

    llama.cpp v0.7.0 New GPU cache for host-memory MoE experts via the --moe-cache-mib option, with experts run on the GPU from an LRU cache for up to 2.2x faster decode; Expert cache is now shared across multiple GPUs; LiliCorr speculative decoding with candidate selection using a s…

RELATED ENTITIES

RELATED TOPICS