You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
I reviewed the Discussions, and have a new and useful enhancement to share.
Feature Description
(Before I start: Llamafile is amazing, thank you all!)
It would be useful - saved me some head-scratching - if the change from the default context size being 512 for every model to being specific by the model was more visible, either in release notes or the README. I can understand why that change was made, but it confused me as to why when I upgraded from 0.8.12 to 0.8.13, models that used to run fine suddenly failed to start.
Motivation
I originally thought it was a bug in 0.8.13, because this command: ./llamafile-0.8.12 -m Meta-Llama-3.1-8B-Instruct.Q4_K_M.gguf works fine whereas ``./llamafile-0.8.13 -m Meta-Llama-3.1-8B-Instruct.Q4_K_M.gguf` gives an allocation error on my M1 MBP.
Specifying the context size with -c resolves the problem easily, and when I look closely at the logs I see that it's precisely because n_ctx is different now. I only wish that this solution was more discoverable. It didn't occur to me that the context size might have changed and I couldn't find any mention of it in the release notes. Hopefully this ticket itself will save some other people time.
Possible Implementation
No response
The text was updated successfully, but these errors were encountered:
Prerequisites
Feature Description
(Before I start: Llamafile is amazing, thank you all!)
It would be useful - saved me some head-scratching - if the change from the default context size being 512 for every model to being specific by the model was more visible, either in release notes or the README. I can understand why that change was made, but it confused me as to why when I upgraded from 0.8.12 to 0.8.13, models that used to run fine suddenly failed to start.
Motivation
I originally thought it was a bug in 0.8.13, because this command:
./llamafile-0.8.12 -m Meta-Llama-3.1-8B-Instruct.Q4_K_M.gguf
works fine whereas ``./llamafile-0.8.13 -m Meta-Llama-3.1-8B-Instruct.Q4_K_M.gguf` gives an allocation error on my M1 MBP.Specifying the context size with
-c
resolves the problem easily, and when I look closely at the logs I see that it's precisely becausen_ctx
is different now. I only wish that this solution was more discoverable. It didn't occur to me that the context size might have changed and I couldn't find any mention of it in the release notes. Hopefully this ticket itself will save some other people time.Possible Implementation
No response
The text was updated successfully, but these errors were encountered: