Instructions to use jsfs11/MixtureofMerges-MoE-4x7b-v3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jsfs11/MixtureofMerges-MoE-4x7b-v3 with Transformers:

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

pipe = pipeline("text-generation", model="jsfs11/MixtureofMerges-MoE-4x7b-v3")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("jsfs11/MixtureofMerges-MoE-4x7b-v3")
model = AutoModelForCausalLM.from_pretrained("jsfs11/MixtureofMerges-MoE-4x7b-v3")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use jsfs11/MixtureofMerges-MoE-4x7b-v3 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jsfs11/MixtureofMerges-MoE-4x7b-v3"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jsfs11/MixtureofMerges-MoE-4x7b-v3",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/jsfs11/MixtureofMerges-MoE-4x7b-v3

SGLang

How to use jsfs11/MixtureofMerges-MoE-4x7b-v3 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 "jsfs11/MixtureofMerges-MoE-4x7b-v3" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jsfs11/MixtureofMerges-MoE-4x7b-v3",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "jsfs11/MixtureofMerges-MoE-4x7b-v3" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jsfs11/MixtureofMerges-MoE-4x7b-v3",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use jsfs11/MixtureofMerges-MoE-4x7b-v3 with Docker Model Runner:
```
docker model run hf.co/jsfs11/MixtureofMerges-MoE-4x7b-v3
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

MixtureofMerges-MoE-4x7b-v3

MixtureofMerges-MoE-4x7b-v3 is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

🧩 Configuration

base_model: senseable/WestLake-7B-v2
gate_mode: hidden
dtype: bfloat16
experts:
  - source_model: jsfs11/RandomMergeNoNormWEIGHTED-7B-DARETIES
    positive_prompts:
      - "Answer this question from the ARC (Argument Reasoning Comprehension)."
      - "Use common sense and logical reasoning skills."
    negative_prompts:
      - "nonsense"
      - "irrational"
      - "math"
      - "code"
  - source_model: senseable/WestLake-7B-v2
    positive_prompts:
      - "Answer this question from the Winogrande test."
      - "Use advanced knowledge of culture and humanity"
    negative_prompts:
      - "ignorance"
      - "uninformed"
      - "creativity"
  - source_model: mlabonne/OmniBeagle-7B
    positive_prompts:
      - "Calculate the answer to this math problem"
      - "My mathematical capabilities are strong, allowing me to handle complex mathematical queries"
      - "solve for"
    negative_prompts:
      - "incorrect"
      - "inaccurate"
      - "creativity"
  - source_model: vanillaOVO/supermario_v3
    positive_prompts:
      - "Predict the most plausible continuation for this scenario."
      - "Demonstrate understanding of everyday commonsense in your response."
      - "Use contextual clues to determine the most likely outcome."
      - "Apply logical reasoning to complete the given narrative."
      - "Infer the most realistic action or event that follows."
    negative_prompts:
      - "guesswork"
      - "irrelevant information"
      - "contradictory response"
      - "illogical conclusion"
      - "ignoring context"

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "jsfs11/MixtureofMerges-MoE-4x7b-v3"

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.	75.31
AI2 Reasoning Challenge (25-Shot)	74.40
HellaSwag (10-Shot)	88.62
MMLU (5-Shot)	64.82
TruthfulQA (0-shot)	70.78
Winogrande (5-shot)	85.00
GSM8k (5-shot)	68.23

Downloads last month: 68

Safetensors

Model size

24B params

Tensor type

BF16

Model tree for jsfs11/MixtureofMerges-MoE-4x7b-v3

jsfs11/RandomMergeNoNormWEIGHTED-7B-DARETIES

mlabonne/OmniBeagle-7B

senseable/WestLake-7B-v2

vanillaOVO/supermario_v3

Merge model

this model

Quantizations

2 models

Evaluation results

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

74.400
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

88.620
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

64.820
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

70.780
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

85.000
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

68.230