RichardErkhov/Ba2han_-_BruinsV2-OpHermesNeu-11B-8bits

Quantization made by Richard Erkhov.

Github

Discord

Request more models

BruinsV2-OpHermesNeu-11B - bnb 8bits

• Model creator: https://huggingface.co/Ba2han/
• Original model: https://huggingface.co/Ba2han/BruinsV2-OpHermesNeu-11B/

Original model description:

license: mit

Task	Version	Metric	Value		Stderr
arc_challenge	0	acc	0.6527	±	0.0139
		acc_norm	0.6869	±	0.0136

Warning! This model may or may not be contaminated See discussion 474. What a shame. It still does perform well though

A passthrough merge of OpenHermes-2.5-neural-chat-7b-v3-1 and Bruins-V2. To be updated.

Template: ChatML

My settings:

Temperature: 0.7-0.8

Min_p: 0.12

Top_K: 0

Repetition Penalty: 1.16

Mirostat Tau: 2.5-3

Mirostat Eta: 0.12

Personal Thoughts:

• The model sometimes throws wrong tags, you can add those to "Custom stopping strings" in Oobabooga.
• Output with Mirostat consistently felt smarter than a set Top_K rate.

Note: The model is hallucinating hard in chat mode for me in some instances, like writing adblocker messages. Kind of funny.

I am not sure which dataset involved was poisoned.