Instructions to use abideen/MonarchCoder-MoE-2x7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use abideen/MonarchCoder-MoE-2x7B with Transformers:

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

pipe = pipeline("text-generation", model="abideen/MonarchCoder-MoE-2x7B")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("abideen/MonarchCoder-MoE-2x7B")
model = AutoModelForCausalLM.from_pretrained("abideen/MonarchCoder-MoE-2x7B")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use abideen/MonarchCoder-MoE-2x7B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "abideen/MonarchCoder-MoE-2x7B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "abideen/MonarchCoder-MoE-2x7B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/abideen/MonarchCoder-MoE-2x7B

SGLang

How to use abideen/MonarchCoder-MoE-2x7B 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 "abideen/MonarchCoder-MoE-2x7B" \
    --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": "abideen/MonarchCoder-MoE-2x7B",
		"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 "abideen/MonarchCoder-MoE-2x7B" \
        --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": "abideen/MonarchCoder-MoE-2x7B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use abideen/MonarchCoder-MoE-2x7B with Docker Model Runner:
```
docker model run hf.co/abideen/MonarchCoder-MoE-2x7B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

MonarchCoder-MoE-2x7B

MonarchCoder-MoE-2x7B is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

The main aim behind creating this model is to create a model that performs well in reasoning, conversation, and coding. AlphaMonarch performs amazing on reasoning and conversation tasks. Merging AlphaMonarch with a coding model yielded MonarchCoder-2x7B which performs better on OpenLLM, Nous, and HumanEval benchmark.

🏆 Evaluation results

|             Metric              |MonarchCoder-Moe-2x7B||MonarchCoder-7B||AlphaMonarch|
|---------------------------------|---------------------|-----------------|------------|
|Avg.                             |       74.23         |      71.17      |   75.99    |
|HumanEval                        |       41.15         |      39.02      |   34.14    |
|HumanEval+                       |       29.87         |      31.70      |   29.26    |
|MBPP                             |       40.60         |       *         |     *      |
|AI2 Reasoning Challenge (25-Shot)|       70.99         |      68.52      |   73.04    |
|HellaSwag (10-Shot)              |       87.99         |      87.30      |   89.18    |
|MMLU (5-Shot)                    |       65.11         |      64.65      |   64.40    |
|TruthfulQA (0-shot)              |       71.25         |      61.21      |   77.91    |
|Winogrande (5-shot)              |       80.66         |      80.19     .|   84.69    |
|GSM8k (5-shot)           .       |       69.37         |      65.13      |   66.72    |

🧩 Configuration

base_model: paulml/OGNO-7B
gate_mode: hidden
dtype: bfloat16
experts:
  - source_model: mlabonne/AlphaMonarch-7B
    positive_prompts:
    - "Mathematics"
    - "Logical Reasoning"
    - "Intelligent Conversations"
    - "Thoughtful Analysis"
    - "Biology"
    - "Medicine"
    - "Problem-solving Dialogue"
    - "Physics"
    - "Emotional intelligence"

    negative_prompts:
    - "History"
    - "Philosophy"
    - "Linguistics"
    - "Literature"
    - "Art and Art History"
    - "Music Theory and Composition"
    - "Performing Arts (Theater, Dance)"

  - source_model: Syed-Hasan-8503/Tess-Coder-7B-Mistral-v1.0
    positive_prompts:
    - "Coding"
    - "Algorithm Design"
    - "Problem Solving"
    - "Software Development"
    - "Computer"
    - "Code Refactoring"
    - "Web development"
    - "Machine learning"
    negative_prompts:
    - "Education"
    - "Law"
    - "Theology and Religious Studies"
    - "Communication Studies"
    - "Business and Management"
    - "Agricultural Sciences"
    - "Nutrition and Food Science"
    - "Sports Science"

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "abideen/MonarchCoder-MoE-2x7B"

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"])