Instructions to use M4-ai/NeuralReyna-Mini-1.8B-v0.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use M4-ai/NeuralReyna-Mini-1.8B-v0.2 with Transformers:

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

pipe = pipeline("text-generation", model="M4-ai/NeuralReyna-Mini-1.8B-v0.2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("M4-ai/NeuralReyna-Mini-1.8B-v0.2")
model = AutoModelForCausalLM.from_pretrained("M4-ai/NeuralReyna-Mini-1.8B-v0.2")
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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use M4-ai/NeuralReyna-Mini-1.8B-v0.2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "M4-ai/NeuralReyna-Mini-1.8B-v0.2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "M4-ai/NeuralReyna-Mini-1.8B-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/M4-ai/NeuralReyna-Mini-1.8B-v0.2

SGLang

How to use M4-ai/NeuralReyna-Mini-1.8B-v0.2 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 "M4-ai/NeuralReyna-Mini-1.8B-v0.2" \
    --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": "M4-ai/NeuralReyna-Mini-1.8B-v0.2",
		"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 "M4-ai/NeuralReyna-Mini-1.8B-v0.2" \
        --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": "M4-ai/NeuralReyna-Mini-1.8B-v0.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use M4-ai/NeuralReyna-Mini-1.8B-v0.2 with Docker Model Runner:
```
docker model run hf.co/M4-ai/NeuralReyna-Mini-1.8B-v0.2
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

NeuralReyna-Mini-1.8B-v0.2

Description

Taken aloobun/Reyna-Mini-1.8B-v0.2 and further fine-tuned it using DPO using the Intel/orca_dpo_pairs dataset.

This model has capabilities in coding, math, science, roleplay, and function calling.

This model was trained on OpenAI's ChatML prompt format.

Evaluation

AGIEval:

GPT4ALL:

Tasks	Version	Filter	Metric	Value		Stderr
arc_challenge	1	none	acc	0.3208	±	0.0136
		none	acc_norm	0.3336	±	0.0138
arc_easy	1	none	acc	0.6035	±	0.0100
		none	acc_norm	0.5833	±	0.0101
boolq	2	none	acc	0.6526	±	0.0083
hellaswag	1	none	acc	0.4556	±	0.0050
		none	acc_norm	0.6076	±	0.0049
openbookqa	1	none	acc	0.2600	±	0.0196
		none	acc_norm	0.3460	±	0.0213
piqa	1	none	acc	0.7236	±	0.0104
		none	acc_norm	0.7307	±	0.0104
winogrande	1	none	acc	0.6062	±	0.0137

Disclaimer

This model may have overfitted to the DPO training data, and may not perform well.

Contributions

Thanks to @aloobun and @Locutusque for their contributions to this model.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	44.85
AI2 Reasoning Challenge (25-Shot)	37.80
HellaSwag (10-Shot)	60.51
MMLU (5-Shot)	45.04
TruthfulQA (0-shot)	37.75
Winogrande (5-shot)	60.93
GSM8k (5-shot)	27.07

Downloads last month: 68

Safetensors

Model size

2B params

Tensor type

F16

Model tree for M4-ai/NeuralReyna-Mini-1.8B-v0.2

Quantizations

4 models

Datasets used to train M4-ai/NeuralReyna-Mini-1.8B-v0.2

Evaluation results

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

37.800
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

60.510
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

45.040
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

37.750
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

60.930
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

27.070