Siyuan Feng

[Process RFC] Empowering New Scoped Module to the Project

15

In this process RFC, We'd like to propose a process to encourage scoped modules and set expectations about what we anticipate in such inclusion. [rendered](https://github.com/Hzfengsy/tvm-rfcs/blob/empowering-new-scoped-module/rfcs/0095-empowering-new-scoped-module.md) [discuss thread](https://discuss.tvm.apache.org/t/process-rfc-empowering-new-scoped-module-to-the-project/13617)

[Bug] Some Printer Bugs

1

In the relax script, the vars with the same name usually are the same var. However, there are two cases that break the rule. ### B1 The binding var and...

[Relax][Runtime] RNNState for Space State Models

2

This commit adds the RNNState class to the Relax VM, similar to the PagedKVCache, for space state models like RWKV and mamba cc @tqchen @MasterJH5574

[Doc] Change img path to remove mxnet dependency

2

Move cat image url to tlcpack/web-data cc @tqchen

Consider move the repo somewhere else

4

Synr is a tvm-related project so far, which helps TVMScript works well. It does a great job to make AST parser stable across different python versions. However, it is not...

Support Qwen2-MoE Architecture

6

[Qwen1.5-MoE-A2.7B-Chat](https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat) is an open-sourced 14B MoE model, based on Qwen2-MoE architecture. It is possible to run on mobile devices. Note that we need to support multi-device TP for the arch...

Comments for "How to Guides"

4

## Wording - [x] `同义词` -> `分类` https://tvm.hyper.ai/docs/how_to/compile/compile_pytorch#查找同义词集名称 - [x] `规约` -> `规约 (reduce)` https://tvm.hyper.ai/docs/how_to/te_schedules/reduction - [x] `交叉线程规约` -> `跨线程规约` - [x] scan: `扫描` -> `线性` - [x] sketch rule:...

## Wording - [x] tiling 循环切分 ~平铺~ - [x] unrolling 循环展开 - [x] high-level language "上层语言" "高级语言" 都可以 - [x] low-level language 建议使用"底层/下层 语言"，尽量避免使用“低级”语言 ### Comments: - [ ] lower...

When reduction axis is small, it's not necessary to use rfactor. This PR updates the gemv rule to use rfactor only when the reduction axis is large enough.