PulseAugur
EN
LIVE 01:00:28

Ollama v0.32.10-rc0 speeds up model prefill performance

By PulseAugur Editorial · [1 sources] ·

Ollama has released version v0.32.10-rc0, introducing optimizations for double-scale NVFP4 models. This update compiles multiply and cast operations into a single kernel, reducing overhead from separate eager ops. Benchmarks on an M5 Max show a performance increase of approximately 6.7% to 7.9% for prefill operations on models like Qwen3.6-27B and muse-glimmer:30b, while speculative decoding performance remains unaffected. AI

IMPACT Improves inference speed for specific model types, potentially benefiting users running local LLMs.

RANK_REASON This is a software release for an open-source tool that improves performance but does not represent a new frontier model or significant industry shift.

Read on Ollama — Releases →

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

Ollama v0.32.10-rc0 speeds up model prefill performance

How we ranked this

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
This is a software release for an open-source tool that improves performance but does not represent a new frontier model or significant industry shift.
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
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
47 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

Full methodology in our editorial standards.

COVERAGE [1]

  1. Ollama — Releases TIER_1 English(EN) · jessegross ·

    v0.32.10-rc0: nn: speed up prefill on double-scale nvfp4 models

    <p>ModelOpt checkpoints apply a float32 global scale to every projection<br /> output on top of the per-group quantization scales. Running the<br /> multiply and the cast back to the activation dtype as separate eager<br /> ops costs an extra kernel launch and a materialized inte…

RELATED ENTITIES

RELATED TOPICS