CUDA: FA support for Deepseek (Ampere or newer) (#13306)

Browse source

* CUDA: FA support for Deepseek (Ampere or newer)

* do loop unrolling via C++ template

...

This commit is contained in:

Johannes Gäßler 2025-05-09 13:34:58 +02:00 • committed by GitHub

parent 27ebfcacba

commit 0cf6725e9f

No known key found for this signature in database

GPG key ID: B5690EEEBB952194

33 changed files with 852 additions and 547 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download patch file Download diff file Expand all files Collapse all files

960

ggml/src/ggml-cuda/fattn-mma-f16.cuh

View file

File diff suppressed because it is too large Load diff