跳到正文
AI NEWS JOURNAL
AI, IN PERSPECTIVE

搜索 AI 资讯

检索标题、摘要与正文,把分散的信息连起来。

按发布时间排序

llama.cpp · b12054

ggml-zdnn:与主机共享缓冲区 (#31481) 签名者:Aaron Teo **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12054/llama-b12054-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64,启用 KleidiAI) [已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12054/llama-b12054-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12054/llama-b12054-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12054/llama-b12054-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/downlo

事件进展与其他报道

llama.cpp · b12053

ggml:在所有 Apple 平台上使用 `vm_allocate` 代替 `posix_memalign` (#31437) * ggml:在 iOS 上使用 `vm_allocate` 代替 `posix_memalign` * ggml:在所有 Apple 平台上使用 vm_allocate,默认运行 test-host-buffer-release **网站:** - **认证:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12053/llama-b12053-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, 启用 KleidiAI) [已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12053/llama-b12053-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12053/llama-b12053-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://

事件进展与其他报道

llama.cpp · b12050

vulkan:改进 RDNA3 和 RDNA4 上的小规模 split-K 选择(#30146)* vulkan:改进 RDNA3 和 RDNA4 上的小规模 split-K 选择 * 通过移除注释代码来清理测试用例 已按要求移除相关测试 **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12050/llama-b12050-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, 启用了 KleidiAI) [已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12050/llama-b12050-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12050/llama-b12050-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.c

事件进展与其他报道

llama.cpp · b12048

sycl:Intel Arc Pro B70s 上的快速 ggml_sycl_comm_exchange() (#30956) * 更新 ggml-sycl.cpp 将 -sm 张量提示评估的每秒吞吐量提升 +19%,而 PPL 保持不变。 * 更新 ggml-sycl.cpp 将首对 wait() 调用替换为 ext_oneapi_submit_barrier()。发现 wait() 调用会导致 MTP 性能下降。 * 更新 ggml-sycl.cpp 添加 comm_exchange 函数指针,以便在初始化时(而非热路径)根据 SYCL_ENABLE_DEFAULT_CONTEXTS=1 设置在 2 缓冲区和 4 缓冲区的 ggml_sycl_comm_exchange 版本之间进行切换。 * 更新 ggml-sycl.cpp 恢复 4 缓冲区方案的后半部分。 * 更新 ggml-sycl.cpp 删除多余空格。 * 更新 ggml-sycl.cpp 在 Windows 上验证完成之前暂不包含 Windows 支持。 **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](h

事件进展与其他报道

llama.cpp · b12047

sycl:dmmv 代码清理 (#31325) * sycl:将 q8_0 重排序的 dmmv 合并到通用模板中 * sycl:对 dmmv 非 K 启动器进行模板化 * sycl:对 dmmv K-量化启动器进行模板化 **网站:** - **认证报告:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12047/llama-b12047-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, 启用了 KleidiAI) [已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12047/llama-b12047-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12047/llama-b12047-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12047/llama

事件进展与其他报道

llama.cpp · b12038

ggml:修复使用 AVX512 BF16 的 clang-cl 构建问题 (#31276) * 在使用 clang-cl 和 avx512bf16 构建时,正确限制 immintrin.h 的包含 * ggml:为 clang-cl 显式添加向量类型转换 **网站:** - **认证报告:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12038/llama-b12038-bin-macos-arm64.tar.gz) - macOS Apple Silicon (arm64, 启用 KleidiAI) [已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel (x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12038/llama-b12038-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12038/llama-b12038-xcframework.zip) **Linux:** - [Ubuntu x64 (CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12038/lla

事件进展与其他报道

llama.cpp · b12032

sycl:尝试使用 PkgConfig 获取 LevelZero 的头文件和库(#31076)**网站:** - **证明材料:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12032/llama-b12032-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12032/llama-b12032-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12032/llama-b12032-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12032/llama-b12032-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12032/)

事件进展与其他报道

llama.cpp · b12029

mtmd:修复 InternVL CLS 令牌顺序(#30886)**网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12029/llama-b12029-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12029/llama-b12029-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12029/llama-b12029-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12029/llama-b12029-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12029/llama-b12029-b

事件进展与其他报道

llama.cpp · b12028

测试:在`test-llama-archs`中启用Gemma 4(#31401)* 测试:在`test-llama-archs`中启用Gemma 4* 测试:在`test-llama-archs`中为Gemma 4使用GQA**网站:** - **认证报告:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12028/llama-b12028-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12028/llama-b12028-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12028/llama-b12028-xcframework.zip)**Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12028/llama-b12028-bin-ubuntu-x64.

事件进展与其他报道

llama.cpp · b12027

llama:为 LFM3 启用张量拆分(#31431) 由 Claude Opus 5.5 协助 **网站:** - **认证:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12027/llama-b12027-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12027/llama-b12027-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12027/llama-b12027-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12027/llama-b12027-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/

事件进展与其他报道

llama.cpp · b12024

正在将 lfm sc 的张量从 3D 重塑为 2D(#30831)**网站:** - **证明材料:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12024/llama-b12024-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12024/llama-b12024-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12024/llama-b12024-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12024/llama-b12024-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12

事件进展与其他报道

llama.cpp · b12022

服务器:修复 /v1/responses 接口以支持 extra_params 参数(#30959)**网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12022/llama-b12022-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12022/llama-b12022-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12022/llama-b12022-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12022/llama-b12022-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/downloa

事件进展与其他报道

llama.cpp · b12021

mtmd:在不添加黑边填充的情况下调整Gemma 4图像的大小(#30922)* mtmd:在不添加黑边填充的情况下调整Gemma 4图像的大小 * 应用@ngxson的建议 --------- 联合作者:Xuan-Son Nguyen **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12021/llama-b12021-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12021/llama-b12021-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12021/llama-b12021-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/do

事件进展与其他报道

llama.cpp · b12020

opencl:添加 fwht(#31434)共同作者:李赫 **网站:** - **证明文件:** - **macOS/iOS:** - [macOS 苹果芯片(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12020/llama-b12020-bin-macos-arm64.tar.gz) - macOS 苹果芯片(arm64,已启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12020/llama-b12020-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12020/llama-b12020-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12020/llama-b12020-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12020/llama-b1202

事件进展与其他报道

llama.cpp · b12019

聊天:在qwen3-coder中以reasoning_format none格式剥离think块(#31380)* 聊天:接受qwen3-coder响应格式中的think块,且采用reasoning_format none格式 由Claude Opus 5.5辅助 * 聊天:提前定义两种qwen3-coder think块变体 **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12019/llama-b12019-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,启用KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12019/llama-b12019-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12019/llama-b12019-xcframework.zip) **Linu

事件进展与其他报道

llama.cpp · b12018

sycl:修复 CPY 中 f32 到 q2_0/q4_0/q5_0 的舍入问题(#31120)**网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12018/llama-b12018-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用 KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12018/llama-b12018-bin-macos-x64.tar.gz) - [iOS XCFramework](https://github.com/ggml-org/llama.cpp/releases/download/b12018/llama-b12018-xcframework.zip) **Linux:** - [Ubuntu x64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12018/llama-b12018-bin-ubuntu-x64.tar.gz) - [Ubuntu arm64(CPU)](https://github.com/ggml-org/llama.cpp/releases/download/b12018/ll)

事件进展与其他报道

b12007

kv-cache:在 GPU 上保留旋转常量(#31073)* kv-cache:在 GPU 上保留旋转常量 由 Codex 协助 * 测试:验证常驻 KV 旋转缓冲区 由 Codex 协助 * kv-cache:记录剩余的多设备旋转相关工作 由 Codex 协助 * kv-cache:支持逐层旋转缓冲区 由 Codex 协助 * kv-cache:修复混合 SWA 旋转输入及混合放置检查 由 Codex 协助 * 模型:在 mistral-large4 和 dflash 中使用逐层旋转 * 测试:删除旋转缓冲区测试 --------- 共同作者:Pascal **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12007/llama-b12007-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用 KleidiAI)[已禁用](ht

事件进展与其他报道

b12005

opencl:添加对q2_0的初始支持(#31340)* opencl:添加q2_0 AOS移动 合著者:Shaofei Qi 协助者:Codex * opencl:添加q2_0 SOA移动与分块矩阵乘法 合著者:Shaofei Qi 协助者:Codex * opencl:为adreno添加q2_0无shuffle的gemm/gemv 协助者:Codex * opencl:为adreno添加q2_0dp4a * opencl:优化adreno的q2_0解码并缩小批处理大小 --------- 合著者:Li He **网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b12005/llama-b12005-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,启用KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/download/b12005/llama-b12005-bin-macos-x64.tar.g

事件进展与其他报道

b11985

metal:更快的 MXFP4 矩阵向量运算,为一元/二元/GLU 提供完整的线程组(#31075)* metal:通过完整线程组加速 MXFP4 矩阵向量内核 采用 fp16 位技巧解码 fp4 值,而非使用线程组查找表,并将打包的半字节数据以对齐的 32 位字加载。该内核不再使用线程组内存。* metal:以完整线程组启动连续的一元和二元运算 连续一元路径以及行广播二元路径曾每个线程组仅使用一个线程。现可为每个线程组最多使用 256 个线程,并对尾部进行边界检查。* metal:将 GLU 行拆分至多个线程组 之前,一条 GLU 行仅在单个线程组上运行,因此单行解码会为整个操作使用单一线程组。现可将每行拆分至最多包含 256 个线程的多个线程组中。**网站:** - **证明文件:** - **macOS/iOS:** - [macOS 苹果芯片

事件进展与其他报道

b11982

hexagon:支持UPSCALE算子(#31331)* hex-upscale:基础放大操作(尚未完全向量化)* hex-upscale:优化最近邻2倍、4倍及双线性4倍放大* hex-upscale:新增对非连续行和维度折叠的支持* hex-upscale:改用vgather而非标量暂存* hex-upscale:将所有内核选择逻辑移至主机端,同时进行注释审查与代码清理* 测试:新增置换后的UPSCALE测试,以覆盖qwen3-vl的图像转文本场景**网站:** - **证明文件:** - **macOS/iOS:** - [macOS Apple Silicon(arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b11982/llama-b11982-bin-macos-arm64.tar.gz) - macOS Apple Silicon(arm64,已启用KleidiAI)[已禁用](https://github.com/ggml-org/llama.cpp/pull/23780) - [macOS Intel(x64)](https://github.com/ggml-org/llama.cpp/releases/downloa

事件进展与其他报道