PulseAugur
中
实时 05:22:23
English(EN) Training Transformers on 'Inference-Only' NPUs: How VKORE Reached 0.9999 Parity

新软件支持在仅推理NPU上训练Transformer

VIDRAFT开发了VKORE,这是一个旨在支持在仅推理NPU上训练Transformer模型的软件层,挑战了行业内将训练仅限于GPU的常规做法。据报道,这项创新在现有NPU硬件上将单用户推理速度提高了3.8倍,服务器吞吐量提高了4.1倍。此外,VKORE支持在这些NPU上进行LoRA训练,与标准的PyTorch精度达到0.9999,速度比同一服务器上的CPU快约170倍,并且在同等硬件上达到了Nvidia B200约54%的吞吐量。 AI

影响 能够更有效地利用现有的NPU硬件进行训练,可能减少对专用GPU基础设施的需求。

排序理由 该项目详细介绍了一种新颖的软件方法,支持在仅推理NPU上进行Transformer训练,包括性能指标和比较。[lever_c_demoted from research: ic=1 ai=1.0]

在 dev.to — LLM tag 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

新软件支持在仅推理NPU上训练Transformer

本文如何被排名

Signal score
31 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该项目详细介绍了一种新颖的软件方法,支持在仅推理NPU上进行Transformer训练,包括性能指标和比较。[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, 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
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. dev.to — LLM tag TIER_1 English(EN) · AI OpenFree ·

    在“仅推理”NPU上训练Transformer:VKORE如何达到0.9999的奇点

    <p><strong>TL;DR:</strong> NPUs are sold as inference-only accelerators. VIDRAFT built VKORE, a software acceleration layer that leaves the chip untouched and (1) raises single-user inference 3.8x and server throughput 4.1x on the same NPU server, and (2) runs real transformer Lo…