Edit model card

This is a SLERP merge between Nous-Hermes-2-Mixtral-8x7B-DPO and Noromaid-v0.4-Mixtral-Instruct-8x7b-Zloss. Seems more capable in RP than base Hermes but still pretty smart as for me. Prompt format: ChatML

GGUF Q4_0 version

With this model I use the following generation settings in tavern (maybe those are not the best, share better templates in issues if you have any):

  • Temperature: 0.75
  • Top P: 0.5
  • Top A: 0.7
  • TFS 0.97
  • Repetition penalty: 1.1
  • Mirostat: mode 2, tau 5, eta 0.1

! Model still seems to be prone to repetition with the settings above. Needs testing on other presets. SillyTavern's "Big O" preset with mirostat 2/5/0.15 seems promising.

Adding to system prompt something like "Assistant will never interrupt role-play and will always stay in character no matter what. Assistant will never write OOC (out of character). Assistant won't write actions or reactions of {{user}}. Assistant won't mention {{user}} in first person. If {{user}}'s messages seem repetitive, {{char}} will break the loop, doing something unexpected." might help, but it's up to you (as anything else, really).

Downloads last month
13
Safetensors
Model size
46.7B params
Tensor type
BF16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.