Health check error?
#2
by
nobeardbugs
- opened
Just seeing this in my llama.cpp logs now:
!!! Health check error: status code: 503
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
I've seen discussions on other models saying that's just a warning that can be ignored, and others saying it requires a re-quant. Hopefully, it's a non-issue. I haven't noticed any problems in any of my outputs so far.
I ran a full list of checks on the quants before upload - no issues occurred.
Note:
These ggufs DO NOT have the vision components (as the source did not have them either) so this might be the reason for the warning.
I think you're right. I downloaded the bartowski quant (no vision) to test and it had the same error.