Fails to load ggml weights

[jessejohnson]

gpt4all fails to load ggml weights (both old and new formats) downloaded for llama.cpp and alpaca.cpp. Is there a way to convert them to a format gpt4all understands, or is this a bug?

[AndriyMulyar]

If you rename the quantized weights to the same string at the the download gpt4all model, it should load. The string is hardcoded in the C++.

[jessejohnson]

That doesn't seem to affect it. It fails to load with the following error llama_model_load: llama_model_load: unknown tensor '}' in model file

I assumed that since this was based off alpaca.cpp it will be compatible with those weights.

[AndriyMulyar]

Ah yeah, whoops. we added a new token to the vocab which is likely is what's breaking it. You'll have to use the original alpaca.cpp executable. Andriy

…

On Tue, Mar 28, 2023, 8:27 PM Jesse Jojo Johnson ***@***.***> wrote: That doesn't seem to affect it. It fails to load with the following error llama_model_load: llama_model_load: unknown tensor '}' in model file I assumed that since this was based off alpaca.cpp it will be compatible with those weights. — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADJ4TBV2QQKMQ6SR6JYPQZ3W6N6YZANCNFSM6AAAAAAWLHOJO4> . You are receiving this because you commented.Message ID: ***@***.***>

[jessejohnson]

Thanks for clarifying @AndriyMulyar !