happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix reference to GEMMA_WEIGHT_T. Refs #198 

PiperOrigin-RevId: 641161403
This commit is contained in:
Jan Wassenberg 2024-06-07 00:53:58 -07:00 committed by Copybara-Service
parent 12707ade80
commit 8dc0e5ea83
1 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
README.md
View File

@ -149,14 +149,15 @@ For the 8-bit switched floating point weights (sfp), run cmake with no options:
cmake -B build cmake -B build
``` ```
**or** if you downloaded bfloat16 weights (any model *without* `-sfp` in the name), **or** if you downloaded bfloat16 weights (any model *without* `-sfp` in the
instead of running cmake with no options as above, run cmake with WEIGHT_TYPE name), instead of running cmake with no options as above, run cmake with
set to [highway's](https://github.com/google/highway) `hwy::bfloat16_t` type GEMMA_WEIGHT_T set to [highway's](https://github.com/google/highway)
(this will be simplified in the future, we recommend using `-sfp` weights `hwy::bfloat16_t` type. We intend to soon support all weight types without
instead of bfloat16 for faster inference): requiring compiler flags. Note that we recommend using `-sfp` weights instead of
bfloat16 for faster inference.
```sh ```sh
cmake -B build -DWEIGHT_TYPE=hwy::bfloat16_t cmake -B build -DGEMMA_WEIGHT_T=hwy::bfloat16_t
``` ```
After running whichever of the above `cmake` invocations that is appropriate for After running whichever of the above `cmake` invocations that is appropriate for