Very confused with the documentation and leveraging llama.cpp features

[TFWol]

I want to use llamafile/llamafiler for FIM completion in vscode and it looks like using llama.cpp features is the best option.

The documentation points to a file with llama.cpp examples, but it's too abstract and doesn't really help with running via commandline.

The command I'm trying to translate from llama.cpp is:

llama-server -m qwen2.5-coder-3b-q8_0.gguf --port 8080 -ngl 99 -fa -ub 1024 -b 1024 -dt 0.1 --ctx-size 0 --cache-reuse 256

The furthest I can get without "command not found" errors is:

./llamafile -m qwen2.5-coder-3b-q8_0.gguf --server -ngl 99 -fa -b 1024 --ctx-size 0

After hours of reading, I have no clue what the equivalents are for
-ub (--ubatch-size), -dt (--defrag-thold), and --cache-reuse 256

Could anyone offer some guidance or advice?

[cjpais]

Llamafile is quite behind in terms of the upstream features. They are likely not supported at this time.

[TFWol]

Got it. Thanks for the clarification.