Instructions to use SteelStorage/Lumosia-v2-MoE-4x10.7 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SteelStorage/Lumosia-v2-MoE-4x10.7 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="SteelStorage/Lumosia-v2-MoE-4x10.7")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SteelStorage/Lumosia-v2-MoE-4x10.7")
model = AutoModelForCausalLM.from_pretrained("SteelStorage/Lumosia-v2-MoE-4x10.7")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use SteelStorage/Lumosia-v2-MoE-4x10.7 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SteelStorage/Lumosia-v2-MoE-4x10.7"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SteelStorage/Lumosia-v2-MoE-4x10.7",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/SteelStorage/Lumosia-v2-MoE-4x10.7

SGLang

How to use SteelStorage/Lumosia-v2-MoE-4x10.7 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "SteelStorage/Lumosia-v2-MoE-4x10.7" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SteelStorage/Lumosia-v2-MoE-4x10.7",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "SteelStorage/Lumosia-v2-MoE-4x10.7" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SteelStorage/Lumosia-v2-MoE-4x10.7",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use SteelStorage/Lumosia-v2-MoE-4x10.7 with Docker Model Runner:
```
docker model run hf.co/SteelStorage/Lumosia-v2-MoE-4x10.7
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Lumosia-v2-MoE-4x10.7

The Lumosia Series upgraded with Lumosia V2.

What's New in Lumosia V2?

Lumosia V2 takes the original vision of being an "all-rounder" and refines it with more nuanced capabilities.

Topic/Prompt Based Approach:

Diverging from the keyword-based approach of its counterpart, Umbra.

Context and Coherence:

With a base context of 8k scrolling window and the ability to maintain coherence up to 16k.

Balanced and Versatile:

The core ethos of Lumosia V2 is balance. It's designed to be your go-to assistant.

Experimentation and User-Centric Development:

Lumosia V2 remains an experimental model, a mosaic of the best-performing Solar models, (selected based on user experience). This version is a testament to the idea that innovation is a journey, not a destination.

Template:

### System:

### USER:{prompt}

### Assistant:

Settings:

Temp: 1.0
min-p: 0.02-0.1

Evals:

Avg:
ARC:
HellaSwag:
MMLU:
T-QA:
Winogrande:
GSM8K:

Examples:

Example 1:

User:

Lumosia:

Example 2:

User:

Lumosia:

🧩 Configuration

yaml
base_model: DopeorNope/SOLARC-M-10.7B
gate_mode: hidden
dtype: bfloat16

experts:
  - source_model: DopeorNope/SOLARC-M-10.7B
    positive_prompts:
    
    negative_prompts:

  - source_model: Sao10K/Fimbulvetr-10.7B-v1 [Updated]
    positive_prompts:
    
    negative_prompts:

  - source_model: jeonsworld/CarbonVillain-en-10.7B-v4 [Updated]
    positive_prompts:
    
    negative_prompts:

  - source_model: kyujinpy/Sakura-SOLAR-Instruct
    positive_prompts:
    
    negative_prompts:

💻 Usage

python
!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "Steelskull/Lumosia-v2-MoE-4x10.7"

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"])

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	73.75
AI2 Reasoning Challenge (25-Shot)	70.39
HellaSwag (10-Shot)	87.87
MMLU (5-Shot)	66.45
TruthfulQA (0-shot)	68.48
Winogrande (5-shot)	84.21
GSM8k (5-shot)	65.13

Downloads last month: 7

Safetensors

Model size

36B params

Tensor type

BF16

Model tree for SteelStorage/Lumosia-v2-MoE-4x10.7

Quantizations

2 models

Evaluation results

normalized accuracy on AI2 Reasoning Challenge (25-Shot)
test set Open LLM Leaderboard

70.390
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

87.870
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

66.450
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

68.480
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

84.210
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

65.130