Prikol

I don't even know anymore

Overview

I have yet to try it UPD: it sucks, bleh

Sometimes mistakes {{user}} for {{char}} and can't think. Other than that, the behavior is similar to the predecessors.

It sometimes gives some funny replies tho, yay!

If you still want to give it a try, here's the cursed text completion preset for cursed models, which makes them somewhat bearable:

https://files.catbox.moe/qr3s64.json

Or this one:

https://files.catbox.moe/97xryh.json

Prompt format: Llama3

Quants

https://huggingface.co/bartowski/Nohobby_L3.3-Prikol-70B-v0.4-GGUF

Merge Details

Step1

base_model: sophosympatheia/Nova-Tempus-70B-v0.2
merge_method: model_stock
dtype: bfloat16
models:
  - model: nbeerbower/Llama-3.1-Nemotron-lorablated-70B
  - model: sophosympatheia/New-Dawn-Llama-3.1-70B-v1.1
tokenizer:
  source: sophosympatheia/Nova-Tempus-70B-v0.2

Step2

models:
  - model: unsloth/DeepSeek-R1-Distill-Llama-70B
  - model: ArliAI/Llama-3.3-70B-ArliAI-RPMax-v1.4
    parameters:
        select_topk:
          - value: [0.18, 0.3, 0.32, 0.38, 0.32, 0.3]
  - model: Nohobby/AbominationSnowPig
    parameters:
        select_topk:
          - value: [0.1, 0.06, 0.05, 0.05, 0.08]
  - model: SicariusSicariiStuff/Negative_LLAMA_70B
    parameters:
        select_topk: 0.17
  - model: mergekit-community/L3.3-L3.1-NewTempusBlated-70B
    parameters:
        select_topk: 0.55
base_model: mergekit-community/L3.3-L3.1-NewTempusBlated-70B
merge_method: sce
parameters:
  int8_mask: true
  rescale: true
  normalize: true
dtype: float32
out_dtype: bfloat16
tokenizer_source: base