Instructions to use Noumaan/phi3-mini-128k-instruct-4bit-quantized with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Noumaan/phi3-mini-128k-instruct-4bit-quantized with Transformers:

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

pipe = pipeline("text-generation", model="Noumaan/phi3-mini-128k-instruct-4bit-quantized", 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("Noumaan/phi3-mini-128k-instruct-4bit-quantized", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("Noumaan/phi3-mini-128k-instruct-4bit-quantized", 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

vLLM

How to use Noumaan/phi3-mini-128k-instruct-4bit-quantized with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Noumaan/phi3-mini-128k-instruct-4bit-quantized"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Noumaan/phi3-mini-128k-instruct-4bit-quantized",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Noumaan/phi3-mini-128k-instruct-4bit-quantized

SGLang

How to use Noumaan/phi3-mini-128k-instruct-4bit-quantized 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 "Noumaan/phi3-mini-128k-instruct-4bit-quantized" \
    --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": "Noumaan/phi3-mini-128k-instruct-4bit-quantized",
		"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 "Noumaan/phi3-mini-128k-instruct-4bit-quantized" \
        --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": "Noumaan/phi3-mini-128k-instruct-4bit-quantized",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Noumaan/phi3-mini-128k-instruct-4bit-quantized with Docker Model Runner:
```
docker model run hf.co/Noumaan/phi3-mini-128k-instruct-4bit-quantized
```

Model Card for phi3-mini-128k-instruct-4bit-quantized

This model is a 4-bit quantized version of the Phi-3-mini-128k-instruct model, optimized for efficient inference while maintaining performance.

Model Details

Model Description

Developed by: [Noumaan]
Model type: Causal Language Model
Language(s) (NLP): English
License: [Original model license]
Finetuned from model: microsoft/Phi-3-mini-128k-instruct

This model is a 4-bit quantized version of the Phi-3-mini-128k-instruct model. It uses the bitsandbytes library for quantization, allowing for reduced memory usage and faster inference times while aiming to maintain most of the original model's performance.

Model Sources

Repository: [Link to your HuggingFace repo]
Original Model: https://huggingface.co/microsoft/Phi-3-mini-128k-instruct

Uses

Direct Use

This model can be used for various natural language processing tasks such as text generation, completion, and question-answering. It's particularly suitable for deployment in resource-constrained environments or for applications requiring faster inference times.

Out-of-Scope Use

This model should not be used for any malicious purposes or to generate harmful content. It's not suitable for tasks requiring extremely high precision or for making critical decisions without human oversight.

Bias, Risks, and Limitations

The quantization process may introduce slight inaccuracies compared to the full-precision model.
This model inherits any biases present in the original Phi-3-mini-128k-instruct model.
The model's outputs should be treated as suggestions or starting points, not as definitive or factual information.

Recommendations

Users should be aware of the quantization's impact on model performance and validate the model's outputs for their specific use case. It's recommended to compare results with the full-precision model for critical applications.

How to Get Started with the Model

## How to Get Started with the Model

This model is pre-quantized to 4-bit precision. You can load and use it directly without additional quantization:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "your-username/phi3-mini-128k-instruct-4bit-quantized"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.float16
)

tokenizer = AutoTokenizer.from_pretrained(model_name)

# Example usage
input_text = "What is the capital of France?"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Downloads last month: 2

Safetensors

Model size

4B params

Tensor type

F32

F16