PulseAugur
EN
LIVE 05:05:28

llama.cpp b11902 speeds up tensor-split model loading with set_tensor_2d

By PulseAugur Editorial · [1 sources] ·

The llama.cpp project has released version b11902, introducing a new function `set_tensor_2d` designed to significantly accelerate the loading of tensor-split models. This optimization involves batching 2D tensor uploads through shared staging slots, allowing for more efficient data transfer. Initial testing on dual Arc Pro B70 GPUs with the GPT-OSS 120B UD-Q4_K_XL model demonstrated a substantial reduction in runtime, from over 260 seconds to approximately 115 seconds, while ensuring data integrity. AI

IMPACT Improves inference performance for users of llama.cpp by accelerating model loading times.

RANK_REASON This is a software update for a specific tool, not a frontier release or significant industry event.

Read on llama.cpp — Releases →

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

llama.cpp b11902 speeds up tensor-split model loading with set_tensor_2d

How we ranked this

Signal score
2 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
This is a software update for a specific tool, not a frontier release or significant industry event.
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, product
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
Same-day
Cluster formed today. Ranking reflects the current source set at time of score.

Full methodology in our editorial standards.

COVERAGE [1]

  1. llama.cpp — Releases TIER_1 Dansk(DA) · Captain-Tripps ·

    b11902: sycl: add set_tensor_2d for faster tensor-split loading (#30358)

    <ul> <li>sycl: add set_tensor_2d for faster tensor-split loading</li> </ul> <p>Pack host rows into chunks of up to 64 MiB when destination rows are<br /> contiguous, avoiding a separate upload call for each row. Keep the<br /> existing per-row copy path for non-contiguous destina…

RELATED ENTITIES

RELATED TOPICS