---
inference: false
license: cc-by-4.0
---

# Model Card

<p align="center">
  <img src="./icon.png" alt="Logo" width="350">
</p>

This is Owlet-Phi-2-Audio.

Owlet is a family of lightweight but powerful multimodal models. 

We provide Owlet-phi-2-audio, which is built upon [SigLIP](https://huggingface.co/google/siglip-so400m-patch14-384) and [Phi-2](https://huggingface.co/microsoft/phi-2) and [Whisper](https://huggingface.co/openai/whisper-small). 
This model supports both audio and visual signals from video data as input, and performs competitevely on the task of Video Question-Answering(QA).
The training procedure and architecture details are publish [here](https://www.phronetic.ai/blogs).

# Quickstart

Here we show a code snippet to show you how to use the model with transformers.
It accepts a mp4 video file, and wav audio file as the input, and generates the answer to the user query.

Before running the snippet, you need to install the following dependencies:

```shell
pip install torch transformers accelerate pillow decord librosa
```

```python
import torch
import transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
from PIL import Image
import warnings
import librosa

# disable some warnings
transformers.logging.set_verbosity_error()
transformers.logging.disable_progress_bar()
warnings.filterwarnings('ignore')

# set device
device = 'cuda'  # or cpu
torch.set_default_device(device)

# create model
print('Loading the model...')
model = AutoModelForCausalLM.from_pretrained(
    'phronetic-ai/owlet-phi-2-audio',
    torch_dtype=torch.float16, # float32 for cpu
    device_map='auto',
    trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained(
    'phronetic-ai/owlet-phi-2-audio',
    trust_remote_code=True)

print('Model loaded. Processing the query...')
# text prompt
prompt = 'What is happening in the video?'
text = f"A chat between a curious user and an artificial intelligence assistant. \
    The assistant gives helpful, detailed, and polite answers to the user's questions. \
        USER: <audio>\n<image>\n{prompt} ASSISTANT:"
input_ids = tokenizer(text, return_tensors='pt').input_ids.to(model.device)

# video and audio file path
video_file_path = '/data/sample_files/sample.mp4'
audio_file_path = '/data/sample_files/sample.wav'
image_tensor, audio_tensor = (tensor.to(model.device, dtype=model.dtype) for tensor in model.process(video_file_path, audio_file_path, model.config))
# passing token indices
IMAGE_TOKEN_INDEX = tokenizer('<image>').input_ids[0]
AUDIO_TOKEN_INDEX = tokenizer('<audio>').input_ids[0]

# generate
output_ids = model.generate(
    input_ids,
    images=image_tensor,
    audio=audio_tensor,
    IMAGE_TOKEN_INDEX=IMAGE_TOKEN_INDEX,
    AUDIO_TOKEN_INDEX=AUDIO_TOKEN_INDEX,
    max_new_tokens=100,
    use_cache=True)[0]

print(f'Response: {tokenizer.decode(output_ids[input_ids.shape[1]:], skip_special_tokens=True).strip()}')


```