Update quantize command to properly suffix input file

[rizenfrmtheashes]

I believe the original code keeps on quantizing the first file, not all files for models that have more than one input quantize file. This update should fix that.

[fermigas]

This is one of the reasons that only the 7B model runs.

Another reason may be line 124 in dalai/index.js:

model: ./models/${req.model || "7B"}/ggml-model-q4_0.bin

It doesn't seem to be aware that the larger models have more than one q4 file.

Also, in the form (index.ejs, 57-60), pressing autocomplete results in this request:

socket.emit('request', {
prompt: input.textContent,
n_predict: 256
})

which does not specify which model, so defaults to 7B.   

[rizenfrmtheashes]

Another reason may be line 124 in dalai/index.js:

model: ./models/${req.model || "7B"}/ggml-model-q4_0.bin

It doesn't seem to be aware that the larger models have more than one q4 file.

Actually this is fine. the underlying llama.cpp package only requires referencing the first model weights file, even if you have multiple, so this should be okay. This is based on my manual testing of the underlying package used, not an e2e test.

As for your latter comment, yeah, it has no way of using anything other than the 7B model when using the quickstart setup. I ended up manually modifying source to solve this problem. I'm not confident enough in my JS skills to make changes across the repo to have that be a selectable option in the webui, and propagating that down as a request to the server, then underlying cpp package call.

[marcuswestin]

@rizenfrmtheashes I implemented that web ui in #16 - would love a code review if you're up for it.

And I just verified that the requested model gets propagated all the way down to the model execution call

[fermigas]

@marcuswestin a lot of great usability fixes/adds in those commits! LGTM