convert : get general.name from model dir, not its parent

[cebtenzzre]

When I converted dl/Open-Orca_Mistral-7b-OpenOrca, convert.py would set general.name to dl. Now it sets it to Open-Orca_Mistral-7b-OpenOrca like convert-hf-to-gguf.py would.

Am I missing something? I don't understand if the previous behavior was intentional. TheBloke's models on HF don't seem to be affected by this... cc @TheBloke in case you have an opinion?

[slaren]

Originally this is from 8f8c28e. My guess is that it was done this way due to the directory structure of the original llama-2 distribution files.

[ggerganov]

My guess is that it was done this way due to the directory structure of the original llama-2 distribution files.

Yes, that was the intention. It's not the smartest logic, so probably it can be improved. But I like to think that convert.py's main purpose is to convert the original models released by Meta. So it should be compatible in some way or another with the directory structure of those models

[cebtenzzre]

My guess is that it was done this way due to the directory structure of the original llama-2 distribution files.

Where do I get these? I know about https://github.com/facebookresearch/llama but their download.sh just downloads files to a folder with a name like "llama-2-7b" (MODEL_PATH) in the current directory (TARGET_FOLDER of ".").

[cebtenzzre]

Could it be clearly explained how the expected folder structure is meant to be created, so that I can verify that convert.py sets general.name correctly in all cases? So far it seems to me that some manual operations are required to result in the name of the parent directory being relevant, Meta checkpoints or otherwise.

[ggerganov]

I have this on my machine:

$ tree llama*
llama
├── 13B
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   └── params.json
├── 30B
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   ├── consolidated.02.pth
│   ├── consolidated.03.pth
│   └── params.json
├── 65B
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   ├── consolidated.02.pth
│   ├── consolidated.03.pth
│   ├── consolidated.04.pth
│   ├── consolidated.05.pth
│   ├── consolidated.06.pth
│   ├── consolidated.07.pth
│   └── params.json
├── 7B
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   └── params.json
├── tokenizer.model
└── tokenizer_checklist.chk
llama2
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.md
├── LICENSE
├── MODEL_CARD.md
├── README.md
├── Responsible-Use-Guide.pdf
├── USE_POLICY.md
├── download.sh
├── example_chat_completion.py
├── example_text_completion.py
├── llama
│   ├── __init__.py
│   ├── generation.py
│   ├── model.py
│   └── tokenizer.py
├── llama-2-13b
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   └── params.json
├── llama-2-13b-chat
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   └── params.json
├── llama-2-70b
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   ├── consolidated.02.pth
│   ├── consolidated.03.pth
│   ├── consolidated.04.pth
│   ├── consolidated.05.pth
│   ├── consolidated.06.pth
│   ├── consolidated.07.pth
│   └── params.json
├── llama-2-70b-chat
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   ├── consolidated.01.pth
│   ├── consolidated.02.pth
│   ├── consolidated.03.pth
│   ├── consolidated.04.pth
│   ├── consolidated.05.pth
│   ├── consolidated.06.pth
│   ├── consolidated.07.pth
│   └── params.json
├── llama-2-7b
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   └── params.json
├── llama-2-7b-chat
│   ├── checklist.chk
│   ├── consolidated.00.pth
│   └── params.json
├── requirements.txt
├── setup.py
├── tokenizer.model
└── tokenizer_checklist.chk

I've forgotten how I downloaded the llama models, but the llama2 were downloaded through the download.sh script.

The v2 support is more important to get right and the v1 is nice to have, but can be ignored as it's quite outdated now. Apologies for this PR getting accidentally ignored - I think the change is good

[mofosyne]

I'm pretty sure it's safe to merge as I was trying to do the same a while ago as well... and gg thinks it's safe to do so based on his directory structure.

Will merge soon unless someone else can see any issue.

(Context: Was going though list of all approved but not yet merged https://github.com/ggerganov/llama.cpp/pulls?q=is%3Apr+is%3Aopen+review%3Aapproved PRs. Do have a check.)