返回全部动态

llama.cpp b10361 修复 EXAONE 4.5 SWA 启用问题

原标题:b10361

llama.cpp Releases一手来源产品发布质量 76

AI 摘要

llama.cpp 发布 b10361 版本,修复了 EXAONE 4.5 模型滑动窗口注意力(SWA)未启用的问题,该问题源于 hparams 加载顺序错误,导致 MTP 头模型被误判。同时修复了元数据加载路径中 TENSOR_SKIP 张量处理问题,并添加了相关测试。该版本提供了多平台二进制下载。

以上摘要由 AI 生成,可能存在误差。事实请以原文为准。

正文节选

<details open> model : fix SWA not being enabled for EXAONE 4.5 (#26848) * model : fix SWA not being enabled for EXAONE 4.5 load_arch_hparams tests `hparams.n_layer() == 64` before LLM_KV_NEXTN_PREDICT_LAYERS has been read. n_layer() returns n_layer_all - n_layer_nextn and n_layer_nextn defaults to 0, so a GGUF carrying the MTP head (block_count=65, nextn=1) evaluates to 65 and the whole SWA block is skipped. The model type switch further down in the same function reads 64, because by then th


发布时间:2026-08-12 01:12
抓取时间:2026-08-12 01:46
来源机构:ggml-org
阅读原文github.com