Instructions to use oopere/martra-phi-3-mini-dpo with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use oopere/martra-phi-3-mini-dpo with Transformers:

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

pipe = pipeline("text-generation", model="oopere/martra-phi-3-mini-dpo", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("oopere/martra-phi-3-mini-dpo", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("oopere/martra-phi-3-mini-dpo", trust_remote_code=True)
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 Settings

vLLM

How to use oopere/martra-phi-3-mini-dpo with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "oopere/martra-phi-3-mini-dpo"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "oopere/martra-phi-3-mini-dpo",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/oopere/martra-phi-3-mini-dpo

SGLang

How to use oopere/martra-phi-3-mini-dpo 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 "oopere/martra-phi-3-mini-dpo" \
    --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": "oopere/martra-phi-3-mini-dpo",
		"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 "oopere/martra-phi-3-mini-dpo" \
        --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": "oopere/martra-phi-3-mini-dpo",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use oopere/martra-phi-3-mini-dpo with Docker Model Runner:
```
docker model run hf.co/oopere/martra-phi-3-mini-dpo
```

You can see the process with instructions for creating the model in the notebook: Aligning_DPO_phi3.ipynb

To create it, we started with the Phi-3-Mini-4K-Instruct model and applied DPO alignment using the distilabel-capybara-dpo-7k-binarized dataset.

Phi-3 is a state-of-the-art model with 3.8 billion parameters that has outperformed other models with 7 billion parameters. The DPO alignment process has produced good results, modifying the model's responses and making them more similar to those in the capybara dataset.

Row in the Dataset:

{'prompt': '<|user|>\nAssist me in calculating 9319357631 plus 595. Numbers and symbols only, please.<|end|>\n<|assistant|>\n', 'chosen': 'The sum of 9319357631 and 595 is 9319358226.<|end|>\n', 'rejected': 'The result of adding 9319357631 and 595 is 9319363626.<|end|>\n'}

Prompt: 3713841893836/4? Limit your response to mathematical expressions and symbols.

Response from the Base model:

To find the result of the division, we can simply divide the given number by 4: $$ \frac{3713841893836}{4} = 928460473459

Response from the fine-tuned model:

3713841893836 ÷ 4 = 928460473459

If you want to see how the model was created, you can check out the repository where the book's notebooks are kept up-to-date.

Downloads last month: 5

Safetensors

Model size

4B params

Tensor type

BF16

oopere
/

martra-phi-3-mini-dpo

Dataset used to train oopere/martra-phi-3-mini-dpo