Can the raw model be removed after it was compressed ?

[negue]

Hey,

new to squish and still new (my success rate so far is... low with a M1 MBP 16GB) in the local llm area as well,

I tried loading my first model (Qwen3-8B) and soon after I got the "disk storage low" warning 😬 , since its the raw + compressed

What my question is, after compressing, can the raw / previous model be deleted ? or is it still needed after

[wesleyscholl]

@negue Thanks for the comment, how did you load the 8B model? Did you use squish covert or download the pre-squished model from the squish-ai huggingface via squish run?

And for your setup, I'd recommend any model with less than 4B parameters. I run llama3.2:3b on my M3 with 16GB RAM and I get responses in 1-3 seconds while leaving plenty of memory for all other applications.

[wesleyscholl]

@negue And you can delete the raw model, squish only uses the compressed model.

[wesleyscholl]

I am working on a feature to automatically remove the raw model after compression/quantization. And my stretch goal is to stream the raw directly into the squished format. That way the raw model is never saved on disk.