# ✨ Sparkfile

A hobbyist build of mozilla-ai/llamafile for DGX Spark.

`v0.10.0`

both CUDA and Vulkan backends packaged.

after download validate backends

```sh
chmod +x ./sparkfile
./sparkfile --gpu nvidia --cli -m your_local_model.gguf --list-devices
./sparkfile --gpu vulkan --cli -m your_local_model.gguf --list-devices
```

*should see both CUDA and Vulkan backends being reported.


basic usage

```sh
./sparkfile --chat -m your_local_model.gguf
```

```sh
./sparkfile --gpu nvidia --chat -m your_local_model.gguf
```

```sh
./sparkfile --gpu vulkan --chat -m your_local_model.gguf
```

```sh
./sparkfile --server ... <llama-server args>
```

for certain reason you are not launching as an executable itself, e.g. wsl2

```sh
sh ./sparkfile <args>
```

Expected Platforms and OSs
CPU: ANY if supported by llamafile
GPU:
 - aarch64_cpu_linux: DGX Spark (GB10), Jetson Thor
 - x86-64_cpu_linux: (needs a seperate `.so` or recompile)
 - x86-64_cpu_win: (needs a seperate `.dll` or recompile)

---
license: apache-2.0
---