Feature Request: P40 KV Tail optimization

未关闭
#118 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
20/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
冷清
技术栈
cpp

调研方向

Reproduce the failure on a P40 with KV tail enabled, using the reported message KV tail layer 3 native operation is unsupported by its planned backend CUDA0; refusing scheduler fallback as the entry point. Trace where backend capability checks reject the KV tail operation and identify which native operations CUDA0 does not support on this GPU. Done means the P40 can initialize and run with KV tail without that refusal, or the unsupported operation is handled through an allowed fallback.

由索引模型根据 Issue 内容生成。

描述

enhancement
Prerequisites
  • I am running the latest code. Mention the version if possible as well.
  • I carefully followed the README.md.
  • I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
  • I reviewed the Discussions, and have a new and useful enhancement to share.
Feature Description

So on P40 gpus right now kv tail doesn't work because of some unsupported ops.

0.04.306.042 E llama_init_from_model: failed to initialize the context: KV tail layer 3 native operation is unsupported by its planned backend CUDA0; refusing scheduler fallback

I don't know enough myself on how all that works but I'd love to have that available for these older gpus for some better performance for tasks that don't quite otherwise fit.

Motivation

Wider GPU support for some older but still useful gpus

Possible Implementation

No response

主要语言
C++
星标
1.2k
派生
80
平均合并
4 天 20 小时
30 天内合并 PR
2

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南,环境需要你自己搭建:先看它的 README,通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

Anbeeld/beellama.cpp 的其他 Issue

查看 Anbeeld/beellama.cpp 的全部 Issue

相似的 Issue

更多 C++ Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。