Spaces:
Running
on
A100
Issue with merges
so, I got this weird issue: whenever I try to make a merge, even the example ones, I get an error about authorization, as if every model in the merge were gated (they are not)
is there any way to fix or workaround this? if its something I am doing wrong, can anyone help me understand the problem?
am I on a time-out of some sort?
here are some lines:
requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/failspy/Meta-Llama-3-8B-Instruct-abliterated-v3/resolve/main/config.json
further:
Repository Not Found for url: https://huggingface.co/failspy/Meta-Llama-3-8B-Instruct-abliterated-v3/resolve/main/config.json.
further:
OSError: failspy/Meta-Llama-3-8B-Instruct-abliterated-v3 is not a local folder and is not a valid model identifier listed on 'https://huggingface.co/models'
thanks in advance!
https://discuss.huggingface.co/t/problem-access-public-model/138206
i found this which seems similar to the issue I am facing, maybe this helps.
But how to pass token this way when using mergekit space?
The "workaround" I can think of is to clone this space and mod the code, then run it with a pro subscription (to cover the hardware needed), which is not practical for me...
What I ended up doing was skipping this space entirely and running it on Google Colab (and Kaggle, which surprisingly has much more permissive use of GPU, but runtime storage is "small")
try one of these, they are limited, but you can get stuff done for free:
LazyMergeKit this is by @maximelabonne
LazyMergeKitPlus is a modified version of LazyMergeKit, by djuna
Copy of LazyMergeKit the one I am using, edited by me, as I am sure you noticed by its name, it is also based on Lazy Merge Kit, but different, really.
I hope this helps someone.
Have a nice day/night, reader!