Triton 编译器发布: gfx950-tutorial-v1.1: gfx950 tutorial pin v1.1

Triton 编译器发布 · 来源更新 2026/07/14 08:40

今日摘要使用自己的 API，仅供个人查看

来源摘要

Rebased onto upstream triton main with the merged warp-pipeline PR #10840 (always-on sched.barrier after each stage's memory ops, SchedGroupMask::non_mem_non_sideeffect). Drops the tutorial's local fence_loads/keep_order commits (superseded by #10840); keeps the out-of-tree LLIR-scheduler / amdgcnas plugin infra. LLVM pin: 62b7cf96 -> 850a2b1.

阅读原始来源

来源: Triton 编译器发布 · 官方来源
来源更新: 2026/07/14 08:40
首次采集: 2026/09/19 12:56

本文为公开信息索引与摘要，详情及后续变化请以原始来源为准。

llama.cpp 发布: b11049

test-llama-archs : generate dummy test vocab (#29084) Assisted-by: pi:llama.cpp/DeepSeek-V4-Flash-Vision-Exp **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/releases/download/b11049/llama-

来源更新 2026/09/19 17:41国际查看原文

算力与芯片llama.cpp 发布官方来源

llama.cpp 发布: b11048

metal : support qwen4exp hc ops (#29000) Add support for the new DSV4 HC op variants used by qwen4exp: - hc_pre with per-element sigmoid gate (gated variant) - hc_post with identity mixing (comb == nullptr) Assisted-by: pi:llama.cpp/Qwen3.8-27B **Website:** -

来源更新 2026/09/19 17:10国际查看原文

算力与芯片量子位 AI 资讯媒体报道

华为汪涛：华为要打造AI算力底座，只做好一颗芯片远远不够

CANN跨过拐点，华为补上软件生态

发布 2026/09/19 16:28国内查看原文

算力与芯片llama.cpp 发布官方来源

llama.cpp 发布: b11047

cuda : fix CUB argsort corruption caused by in-place keys (#28389) argsort_f32_i32_cuda_cub called the one-shot DeviceRadixSort::SortPairs API with d_keys_in == d_keys_out (temp_keys, temp_keys). CUB's internal double-buffer ping-pong requires distinct key buf

来源更新 2026/09/19 15:27国际查看原文

算力与芯片llama.cpp 发布官方来源

llama.cpp 发布: b11046

opencl: add support for bin kernel `flash_attn_f32_f16_bin` (#29046) * opencl: add `flash_attn_f32_f16_bin` * opencl: guarded prefill fa **Website:** - **Attestations:** - **macOS/iOS:** - [macOS Apple Silicon (arm64)](https://github.com/ggml-org/llama.cpp/rel

来源更新 2026/09/19 07:58国际查看原文