Hugging Face's logo

NeverSleep
/
Llama-3-Lumimaid-8B-v0.1

Text Generation
Transformers
Safetensors
llama
Not-For-All-Audiences
nsfw
conversational
Inference Endpoints
text-generation-inference
Model card Files Community
2

EoS still isn't corrected to 128009?

#2
by Pyroserenus - opened
Discussion
Pyroserenus
•
edited 21 days ago

Model is using the EoS of llama 3 base, not llama 3 instruct. This results in <|eot_id|> leakage on some UI and engine combinations.
image.png

image.png

Undi95
 NeverSleep org

Corrected

Sign up or log in to comment