---
license: apache-2.0
tags:
- moe
- frankenmoe
- merge
- mergekit
- lazymergekit
- Himitsui/Kaiju-11B
- Sao10K/Fimbulvetr-11B-v2
- decapoda-research/Antares-11b-v2
- beberik/Nyxene-v3-11B
base_model:
- Himitsui/Kaiju-11B
- Sao10K/Fimbulvetr-11B-v2
- decapoda-research/Antares-11b-v2
- beberik/Nyxene-v3-11B
---

# Umbra-v3-MoE-4x11b

Umbra-v3-MoE-4x11b is a Mixure of Experts (MoE) made with the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [Himitsui/Kaiju-11B](https://huggingface.co/Himitsui/Kaiju-11B)
* [Sao10K/Fimbulvetr-11B-v2](https://huggingface.co/Sao10K/Fimbulvetr-11B-v2)
* [decapoda-research/Antares-11b-v2](https://huggingface.co/decapoda-research/Antares-11b-v2)
* [beberik/Nyxene-v3-11B](https://huggingface.co/beberik/Nyxene-v3-11B)

## 🧩 Configuration

```yaml
base_model: vicgalle/CarbonBeagle-11B-truthy
gate_mode: hidden
dtype: bfloat16
experts_per_token: 4
experts:
  - source_model: Himitsui/Kaiju-11B
    positive_prompts:
      - "Imagine"
      - "Create"
      - "Envision"
      - "Fantasize"
      - "Invent"
      - "Narrate"
      - "Plot"
      - "Portray"
      - "Storytell"
      - "Visualize"
      - "Describe"
      - "Develop"
      - "Forge"
      - "Craft"
      - "Conceptualize"
      - "Dream"
      - "Concoct"
      - "Characterize"
    negative_prompts:
      - "Recite"
      - "Report"
      - "Summarize"
      - "Enumerate"
      - "List"
      - "Cite"

  - source_model: Sao10K/Fimbulvetr-11B-v2
    positive_prompts:
      - "Dramatize"
      - "Embody"
      - "Illustrate"
      - "Perform"
      - "Roleplay"
      - "Simulate"
      - "Stage"
      - "Unfold"
      - "Weave"
      - "Design"
      - "Outline"
      - "Script"
      - "Sketch"
      - "Spin"
      - "Depict"
      - "Render"
      - "Fashion"
      - "Conceive"
    negative_prompts:
      - "Analyze"
      - "Critique"
      - "Dissect"
      - "Explain"
      - "Clarify"
      - "Interpret"

  - source_model: decapoda-research/Antares-11b-v2
    positive_prompts:
      - "Solve"
      - "Respond"
      - "Convey"
      - "Disclose"
      - "Expound"
      - "Narrate"
      - "Present"
      - "Reveal"
      - "Specify"
      - "Uncover"
      - "Decode"
      - "Examine"
      - "Report"
      - "Survey"
      - "Validate"
      - "Verify"
      - "Question"
      - "Query"
    negative_prompts:
      - "Divert"
      - "Obscure"
      - "Overstate"
      - "Undermine"
      - "Misinterpret"
      - "Skew"

  - source_model: beberik/Nyxene-v3-11B
    positive_prompts:
      - "Explain"
      - "Instruct"
      - "Clarify"
      - "Educate"
      - "Guide"
      - "Inform"
      - "Teach"
      - "Detail"
      - "Elaborate"
      - "Enlighten"
      - "Advise"
      - "Interpret"
      - "Analyze"
      - "Define"
      - "Demonstrate"
      - "Illustrate"
      - "Simplify"
      - "Summarize"
    negative_prompts:
      - "Speculate"
      - "Fabricate"
      - "Exaggerate"
      - "Mislead"
      - "Confuse"
      - "Distort"

```

## 💻 Usage

```python
!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Steelskull/Umbra-v3-MoE-4x11b"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)

messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
```