Refactor: convert_hf_to_gguf.py

[pwilkin]

From the creators of the horror show "llama-model.cpp - refactoring" comes the sequel: "convert_hf_to_gguf.py - refactoring".

Still in draft because it's not all finished, but I thought that if someone wanted to chime in I'll put it up already.

[CISC]

🤯 Looks like a decent enough plan though. Only downside is that there will no longer just be a single script to grab and run, but that has been a source of error and confusion anyway.

You should derive _MODEL_MODULES from models_map.

[CISC]

@compilade Your input would be much appreciated as well.

[compilade]

(note for later) This will (trivially) conflict with at least

• convert : handle compressed-tensors quant method #17069
• convert : parse safetensors directly #15667
• convert : use reflinks for faster conversion #15727
• (non-existent yet, but wip) convert : generalized repacking for pre-quantized models

[CISC]

• convert : handle compressed-tensors quant method #17069
• convert : parse safetensors directly #15667

Are you expecting more work on these? Can be merged now.

[pwilkin]

Okay, so I've actually gone over and verified the conversion results - refactored, applied some tips from here while doing that and most of all - tested that it actually works.

[pwilkin]

Conversion verification script: https://gist.github.com/pwilkin/1e488423e9f2549c0518179bb9f752d5

[pwilkin]

@CISC have fun! :)

[CISC]

@CISC have fun! :)

You need to fix the CIs first, esp. convert_hf_to_gguf_update.py. :)

[pwilkin]

Yeah, might've went out of draft too early :>