| commit | 8a5baef37dabc9160611c24424b633c2e7a6f1b3 | [log] [tgz] |
|---|---|---|
| author | Lei Zhang <antiagainst@google.com> | Wed May 10 18:18:41 2023 -0700 |
| committer | GitHub <noreply@github.com> | Thu May 11 01:18:41 2023 +0000 |
| tree | 64ff1fecd28be5a654aa55502d05f17fa1de17c2 | |
| parent | 5cbcf08eeffac440c2c0bae7daf39a1f442ba41e [diff] |
[stream] Switch to perform tight packing of sub-byte elements (#13344) Thus far, for sub-byte element types, we unconditionally expand them to byte-aligned ones. It wastes storage and memory. This becomes more difficult with recent large language models, where we have a large amount of data for weights. Thus, this commit starts to perform tight packing of sub-byte elements at the stream level. With this, sub-byte element types can flow down the compilation pipeline, allowing each backend to handle it accordingly. Given handling of sub-byte elements are tricky, right now we only enable selected sub-byte element types that is some power of two to get started. Other cases can be enabled when we have the use case. Progress towards https://github.com/openxla/iree/issues/12859
IREE (Intermediate Representation Execution Environment, pronounced as “eerie”) is an MLIR-based end-to-end compiler and runtime that lowers Machine Learning (ML) models to a unified IR that scales up to meet the needs of the datacenter and down to satisfy the constraints and special considerations of mobile and edge deployments.
See our website for project details, user guides, and instructions on building from source.
IREE is still in its early phase. We have settled down on the overarching infrastructure and are actively improving various software components as well as project logistics. It is still quite far from ready for everyday use and is made available without any support at the moment. With that said, we welcome any kind of feedback on any communication channels!
See our website for more information.
IREE is licensed under the terms of the Apache 2.0 License with LLVM Exceptions. See LICENSE for more information.