PulseAugur
中
实时 20:45:33
English(EN) b12487: Title: CUDA: few-row MMQ for MUL_MAT_ID (#32065)

llama.cpp 提升 CUDA 性能并修复 AMD 设备错误

llama.cpp 项目发布了专注于 CUDA 性能提升和错误修复的更新。版本 b12487 在 DGX Spark 上为 MUL_MAT_ID 引入了少行矩阵乘法优化,提高了效率。版本 b12433 解决了 AMD 设备上与 MUL_MAT_ID 相关的特定中止问题,该问题发生在处理奇数列源和某些浮点格式时,确保了跨不同硬件配置的稳定性。 AI

影响 llama.cpp 的性能改进和错误修复可以提高本地 AI 模型执行的效率。

排序理由 开源项目的软件发布说明,专注于性能优化和错误修复。

在 llama.cpp — Releases 阅读 →

AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →

llama.cpp 提升 CUDA 性能并修复 AMD 设备错误

本文如何被排名

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
开源项目的软件发布说明,专注于性能优化和错误修复。
Source corroboration
3 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
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
6 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.
Coverage growth since scoring
+1 source(s) since last score
New sources have picked up this story since our last re-score. Score will update on the next scoring pass.

完整方法见我们的编辑标准。

报道来源 [3]

  1. llama.cpp — Releases TIER_1 English(EN) · am17an ·

    b12533: CUDA: 在 mul_mat_q 中使用 64 位 dst 偏移量 (#32374)

    <p>The MMQ kernels computed the dst tile offset (jt<em>J</em>stride_col_dst) in<br /> 32-bit int. For a dst with more than 2^31 elements, column tiles that<br /> start past INT_MAX/nrows wrap around and write out of bounds.</p> <p>This happens with an all-logits output head, e.g.…

  2. llama.cpp — Releases TIER_1 English(EN) · anujj ·

    b12487: 标题:CUDA:MUL_MAT_ID 的几行 MMQ (#32065)

    <ul> <li> <p>CUDA: few-row MMQ for MUL_MAT_ID</p> </li> <li> <p>CUDA: few-row MMQ for MUL_MAT_ID from 2 tokens on DGX Spark</p> </li> <li> <p>CUDA: few-row MUL_MAT_ID: one launcher, tighter pair count, more tests</p> </li> <li> <p>CUDA: few-row MUL_MAT_ID: dispatch order MMVQ, fe…

  3. llama.cpp — Releases TIER_1 English(EN) · liminfei-amd ·

    b12433: cuda : 修复 AMD 上奇数列 src0 时的 MUL_MAT_ID 中止 (#32385)

    <p>On AMD devices MUL_MAT_ID with an F32, F16 or BF16 src0 and at most 8 tokens always used MMVF.<br /> MMVF reads src0 in pairs of values and asserts an even ne[0], so a rank 1 LoRA on MoE expert weights (lora_b has ne[0] == 1) aborted.<br /> Use MMVF there only for an even ne[0…