Update README.md
Browse files
README.md
CHANGED
|
@@ -73,6 +73,16 @@ full 128k size. See our
|
|
| 73 |
repository for llamafiles that are known to work with a 128kb context
|
| 74 |
size.
|
| 75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
On GPUs with sufficient RAM, the `-ngl 999` flag may be passed to use
|
| 77 |
the system's NVIDIA or AMD GPU(s). On Windows, only the graphics card
|
| 78 |
driver needs to be installed. If the prebuilt DSOs should fail, the CUDA
|
|
|
|
| 73 |
repository for llamafiles that are known to work with a 128kb context
|
| 74 |
size.
|
| 75 |
|
| 76 |
+
On Windows there's a 4GB limit on executable sizes. You can work around
|
| 77 |
+
that by downloading the [official llamafile
|
| 78 |
+
release](https://github.com/Mozilla-Ocho/llamafile/releases) binary,
|
| 79 |
+
renaming it to have a .exe extension, and then passing the llamafiles in
|
| 80 |
+
this repo via the `-m` flag as though they were GGUF weights, e.g.
|
| 81 |
+
|
| 82 |
+
```
|
| 83 |
+
.\llamafile-0.8.11.exe -m Meta-Llama-3.1-405B.Q2_K.llamafile
|
| 84 |
+
```
|
| 85 |
+
|
| 86 |
On GPUs with sufficient RAM, the `-ngl 999` flag may be passed to use
|
| 87 |
the system's NVIDIA or AMD GPU(s). On Windows, only the graphics card
|
| 88 |
driver needs to be installed. If the prebuilt DSOs should fail, the CUDA
|