README.md

metadata

license: apache-2.0
tags:
  - moe
  - frankenmoe
  - merge
  - mergekit
  - lazymergekit

Lumina-RP

Lumina-4 is a Mixture of Experts (MoE) made with LazyMergekit. This model has improved roleplaying and storytelling from Lumina-3.5 while still retaining its strength. A slight improvement on the Open LLM leaderboard. It uses a context window of up to 32k.

Response Examples

Input: 25 - 4 * 2 + 3 = ?
------------------------------------
Output: The given mathematical expression is 25 - 4 * 2 + 3. To solve it, we should follow the order of operations (PEMDAS):

1. Parentheses: There are none in this equation.
2. Exponents: None present.
3. Multiplication and Division: We perform multiplications before divisions from left to right. In this case, 4 * 2 = 8. Now our equation becomes 25 - 8 + 3.
4. Addition and Subtraction: We perform additions before subtractions from left to right. Here, we have a subtraction then an addition: 25 - 8 = 17, then 17 + 3 = 20.

So the result of the expression 25 - 4 * 2 + 3 is 20.

🏆 Open LLM Leaderboard Evaluation Results

Metric	Value
Avg.	75.59
AI2 Reasoning Challenge (25-Shot)	72.61
HellaSwag (10-Shot)	88.45
MMLU (5-Shot)	64.73
TruthfulQA (0-shot)	74.24
Winogrande (5-shot)	83.90
GSM8k (5-shot)	69.60

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Ppoyaa/Lumina-4"

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"])