---
license: cc-by-nc-4.0
base_model: Qwen/Qwen2-7B-Instruct
model-index:
- name: Dolphin 
  results: []
tags:
- RAG
- on-device language model
- Retrieval Augmented Generation
inference: false
space: false
spaces: false
language:
- en
---
# Dolphin: Long Context as a New Modality for on-device RAG

<p align="center">
- <a href="https://www.nexaai.com/models" target="_blank">Nexa Model Hub</a>
- <a href="https://arxiv.org/pdf/2408.15518" target="_blank">ArXiv</a>
</p>

<p align="center" width="100%">
  <a><img src="logo.png" alt="nexa-octopus" style="width: 30%; min-width: 300px; display: block; margin: auto;"></a>
</p>

## Overview
Dolphin is a novel approach to accelerate language model inference by treating long context as a new modality, similar to image, audio, and video modalities in vision-language models. This innovative method incorporates a language encoder model to encode context information into embeddings, applying multimodal model concepts to enhance the efficiency of language model inference。 Below are model highlights:
- 🧠 Context as a distinct modality
- 🗜️ Language encoder for context compression
- 🔗 Multimodal techniques applied to language processing
- ⚡ Optimized for energy efficiency and on-device use
- 📜 Specialized for long context understanding

## Model Architecture
Dolphin employs a decoder-decoder framework with two main components:
1. A smaller decoder (0.5B parameters) for transforming information from extensive contexts
2. A larger decoder (7B parameters) for comprehending and generating responses to current queries
3. The architecture also includes a projector to align embeddings between the text encoder and the main decoder.

![Model Architecture](modelstructure.jpg)

## Running the Model
### Method 1
download this repository and run the following commands:
```bash
git lfs install
git clone https://huggingface.co/NexaAIDev/Dolphin
python inference_example.py
```

### Method 2
Install `nexaai-dolphin` package
```
pip install nexaai-dolphin
```

Then run the following commands:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer, AutoConfig
import torch
from dolphin.configuration_dolphin import DolphinConfig
from dolphin.modeling_dolphin import DolphinForCausalLM


def inference_instruct(mycontext, question, device="cuda:0"):
    import time
    MEMORY_SIZE = 32
    start_time = time.time()
    generated_token_ids = []
    prompt = f" <context>{question}"
    text_chunks = [tokenizer(chunk).input_ids for chunk in prompt.split("<context>")]
    input_ids = (
        torch.tensor(
            text_chunks[0] + [-1] * MEMORY_SIZE + text_chunks[1], dtype=torch.long
        )
        .unsqueeze(0)
        .to(device)
    )
    context_tokenized = tokenizer(
        mycontext + "".join([f"[memory_{i}]" for i in range(MEMORY_SIZE)]),
        return_tensors="pt",
    )
    context_tokenized = {k: v.to(device) for k, v in context_tokenized.items()}
    context_token_count = (context_tokenized["input_ids"]).shape[1] - MEMORY_SIZE
    for i in range(context_token_count):
        next_token = (
            model(
                input_ids,
                context_input_ids=context_tokenized["input_ids"],
                context_attention_mask=context_tokenized["attention_mask"],
            )
            .logits[:, -1]
            .argmax(-1)
        )
        if next_token.item() == 151643:
            break
        generated_token_ids.append(next_token.item())
        input_ids = torch.cat([input_ids, next_token.unsqueeze(1)], dim=-1)
    result = tokenizer.decode(generated_token_ids)
    print(f"Time taken: {time.time() - start_time}")
    return result


if __name__ == "__main__":
    device_name = "cuda:0" if torch.cuda.is_available() else "cpu"
    AutoConfig.register("dolphin", DolphinConfig)
    AutoModelForCausalLM.register(DolphinConfig, DolphinForCausalLM)
    tokenizer = AutoTokenizer.from_pretrained('NexaAIDev/Dolphin')
    model = AutoModelForCausalLM.from_pretrained('NexaAIDev/Dolphin', trust_remote_code=True, torch_dtype=torch.bfloat16, device_map=device_name)
    
    # Run inference example
    mycontext = "Nexa AI is a Cupertino-based company founded in May 2023 that researches and develops models and tools for on-device AI applications. The company is founded by Alex and Zack. The company is known for its Octopus-series models, which rival large-scale language models in capabilities such as function-calling, multimodality, and action-planning, while remaining efficient and compact for edge device deployment. Nexa AI's mission is to advance on-device AI in collaboration with the global developer community. To this end, the company has created an on-device model hub for users to find, share, and collaborate on open-source AI models optimized for edge devices, as well as an SDK for developers to run and deploy AI models locally"
    question = "Who founded Nexa AI?"
    result = inference_instruct(mycontext, question, device=device_name)
    print("Result:", result)
```

## Training Process
Dolphin's training involves three stages:
1. Restoration Training: Reconstructing original context from compressed embeddings
2. Continual Training: Generating context continuations from partial compressed contexts
3. Instruction Fine-tuning: Generating responses to queries given compressed contexts

This multi-stage approach progressively enhances the model's ability to handle long contexts and generate appropriate responses.

## Citation
If you use Dolphin in your research, please cite our paper:

```bibtex
@article{chen2024dolphinlongcontextnew,
      title={Dolphin: Long Context as a New Modality for Energy-Efficient On-Device Language Models}, 
      author={Wei Chen and Zhiyuan Li and Shuo Xin and Yihao Wang},
      year={2024},
      eprint={2408.15518},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2408.15518}, 
}
```

## Contact
For questions or feedback, please [contact us](octopus@nexa4ai.com)