llama.cpp b10684 发布：改进 SYCL 显存适配算法

AI 摘要

llama.cpp 发布 b10684 版本，改进了 SYCL 后端中 --fit 算法对 --fit-target 的遵循，更准确地计算给定上下文大小所需的峰值显存，避免 OOM 并减少过度保留。在 Arc b70 上测试，使用 unsloth 的 qwen3.8（Q4_K_XL）模型，配合 q8_0 KV 和 MTP，在 --fit-target 1 下可实现 262144 上下文完全可用。

正文节选

<details open> sycl: make --fit respect --fit-target better (#27629) improve the --fit algorithm to take into account the actual peak required VRAM for a given context size on a SYCL backend. This includes both properly accounting for how much VRAM is required when the allocated context is fully used (which makes the reported context drop below what it did before, but stop it OOMing) as well as preventing some overly-conservative calculations which meant too much VRAM was being reserved. Tes

发布时间：2026-08-29 23:57

抓取时间：2026-08-30 00:16

来源机构：ggml-org