Overview

Request 1201559 accepted

- update to version 4.4.0
**Removed**: Flash Attention support in the Python package due to
significant package size increase with minimal
performance gain.
### New features
* Support Llama3
* Support Gemma2
* Add log probs for all tokens in vocab
* Grouped conv1d

### Fixes and improvements
* Some improvements in flash attention
* Fix crash when using return_alternative on CUDA
* Quantization AWQ GEMM + GEMV

Loading...
Request History
Adrian Schröter's avatar

adrianSuSE created request

- update to version 4.4.0
**Removed**: Flash Attention support in the Python package due to
significant package size increase with minimal
performance gain.
### New features
* Support Llama3
* Support Gemma2
* Add log probs for all tokens in vocab
* Grouped conv1d

### Fixes and improvements
* Some improvements in flash attention
* Fix crash when using return_alternative on CUDA
* Quantization AWQ GEMM + GEMV


Ana Guerrero's avatar

anag+factory added openSUSE:Factory:Staging:adi:24 as a reviewer

Being evaluated by staging project "openSUSE:Factory:Staging:adi:24"


Ana Guerrero's avatar

anag+factory accepted review

Picked "openSUSE:Factory:Staging:adi:24"


Factory Auto's avatar

factory-auto added opensuse-review-team as a reviewer

Please review sources


Factory Auto's avatar

factory-auto accepted review

Check script succeeded


Marcus Rueckert's avatar

darix accepted review

Accepted review for by_group opensuse-review-team request 1201559 from user factory-auto


Saul Goodman's avatar

licensedigger accepted review

The legal review is accepted preliminary. The package may require actions later on.


Ana Guerrero's avatar

anag+factory accepted review

Staging Project openSUSE:Factory:Staging:adi:24 got accepted.


Ana Guerrero's avatar

anag+factory approved review

Staging Project openSUSE:Factory:Staging:adi:24 got accepted.


Ana Guerrero's avatar

anag+factory accepted request

Staging Project openSUSE:Factory:Staging:adi:24 got accepted.

openSUSE Build Service is sponsored by