PulseAugur
EN
LIVE 17:06:07

NUMALLAMA update boosts ARM server LLM speeds, adds Hugging Face models

The NUMALLAMA project has released an update that significantly boosts prompt processing speeds, particularly for Llama 3.1 8B Q8 models, which now achieve around 800 tokens/sec. This update also introduces ready-made models on Hugging Face, including Qwen3-30B-A3B, Qwen3.8-27B, and Qwen2.5, compatible with OpenAI clients. The performance optimizations are specifically tuned for ARM64 architecture, with notable speed improvements observed on dual-socket ARM servers without GPUs. AI

IMPACT Improves LLM performance on ARM hardware, making it more accessible for users without GPUs.

RANK_REASON This is an update to an open-source project that improves performance and accessibility of LLMs on specific hardware, rather than a new frontier model release.

Read on dev.to — LLM tag →

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

NUMALLAMA update boosts ARM server LLM speeds, adds Hugging Face models

How we ranked this

Signal score
36 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
This is an update to an open-source project that improves performance and accessibility of LLMs on specific hardware, rather than a new frontier model release.
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. dev.to — LLM tag TIER_1 English(EN) · Miloserdev ·

    NUMALLAMA update: Qwen3-30B-A3B at ~90 tok/s on a dual-socket ARM server, no GPU

    <p>What changed since last time</p> <ul> <li>Faster prompt processing. Activation packing borrowed from Arm KleidiAI’s layout, balanced work split across threads, and larger prefill blocks: Llama 3.1 8B Q8 went from ~620 to ~800 tok/s. The output is bit-identical to before.</li> …