Awesome

A train-less, much faster decoding algorithm

Ouroboros is a new decoding algorithm that has much faster generation speed compared with Speculative Decoding and Lookahead Decoding, without any training cost.

Here is an example of generation with prompt

Write a Java program to implement a heap
class Heap {   

Blue words are generated by small model, black words are corrected by large model and red words are accepted candidate suffixes.

Benchmark Result

The generation speed tested on HumanEval and MBPP are presented here. The shadowed parts are the extra improvement caused by warm start.

Method

We utilize phrases to enhance the efficiency of the draft model in Speculative Sampling. These phrases are derived from the following sources:

1. Lookahead Decoding: This method generates phrases concurrently with the forward pass of the draft model.
2. Verification: Useful phrases identified during the verification phase of Speculative Sampling.
3. Historical Contexts: Phrases generated from similar user prompts or similar tasks are leveraged.

These phrases are then employed enhance the efficiency of the draft model by 2 ways, without affecting the draft quality:

1. By enabling parallelization of the drafting process.
2. By increasing the draft length, thereby generating multiple candidate drafts.

Install

Install from pip

pip install ourodec

Install from source code

• First, clone this repo as ouroboros
• Then, run the following command

cd ouroboros
pip install -e .

Quick Start

An Example

• Here is an example of using Ouroboros. Note: current version of ouroboros can only support Llama model inference. Please import LlamaForCausalLM from ouroboros.models instead of transformers

• The following code shows an example of comparing the output of Ouroboros with Greedy decoding. If ouroboros is installed correctly, the following code should output True.

import torch
from ouroboros import ouroboros
from transformers import AutoTokenizer
from ouroboros.models import LlamaForCausalLM

window_size = 20
guess_set_size = 20
lookahead_level = 7
gamma = 12

small_model = LlamaForCausalLM.from_pretrained("yourpath", torch_dtype=torch.float16, device_map='cuda')
target_model = LlamaForCausalLM.from_pretrained("yourpath", torch_dtype=torch.float16, device_map='cuda')

tokenizer = AutoTokenizer.from_pretrained("yourpath")

prompt = "Please summarize the following paragraph. Officers searched properties in the Waterfront Park and Colonsay View areas of the city on Wednesday. Detectives said three firearms, ammunition and a five-figure sum of money were recovered. A 26-year-old man who was arrested and charged appeared at Edinburgh Sheriff Court on Thursday. Summary: "

input_ids = tokenizer(prompt, return_tensors='pt').to('cuda')['input_ids']

ouroboros_output = ouroboros(input_ids, small_model, target_model, max_len=64, gamma=gamma, window_size=window_size, guess_set_size=guess_set_size, lookahead_level=lookahead_level)

std_output = target_model.generate(input_ids, do_sample=False, min_length=64, max_length=64)

print(ouroboros_output[:,:64].equal(std_output[:,:64]))

Using Ouroboros

• First, prepare two Hugging Face transformers models (the target model and draft model). Load the model with LlamaForCausalLM from ouroboros.models.
• Then import the generation function ouroboros from ouroboros.
• Call the function by the following parameters:

@torch.no_grad()
def ouroboros(prefix : torch.Tensor, approx_model : torch.nn.Module, target_model : torch.nn.Module, ngram_cache : CacheEngine = None,
                         max_len : int = 512 , gamma : int = 4, window_size = 20, guess_set_size = 20, lookahead_level = 7, eos_token_id = 2, topk=3) -> torch.Tensor:
    """
    Performs ouroboros with an approximate model and a target model to generate a sequence of tokens.

    Args:
        prefix (torch.Tensor): The initial sequence of tokens to start the generation from.
        approx_model (torch.nn.Module): The approximate model used for initial token generation. The model should support Hugging Face transformers model methods.
        target_model (torch.nn.Module): The target model used for refining the generated tokens. The model should support Hugging Face transformers model methods.
        ngram_cache (CacheEngine, optional): A cache engine for storing and retrieving n-gram predictions. Defaults to None, in which case a new cache engine is created.
        max_len (int, optional): The maximum length of the generated sequence. Defaults to 512.
        gamma (int, optional): The lookahead parameter for generation. Defaults to 4.
        window_size (int, optional): The window size used for n-gram generation. Defaults to 20. Currently, must be equal to guess_set_size.
        guess_set_size (int, optional): The size of the guess set for n-gram retrieving. Defaults to 20. Currently, must be equal to window_size.
        lookahead_level (int, optional): The level of lookahead decoding. Defaults to 7.
        eos_token_id (int, optional): The token id representing the end-of-sequence token. Defaults to 2. Should be given by tokenizer.eos_token_id.
        topk (int, optional): The k of verifying k candidate suffixes.

    Returns:
        torch.Tensor: The generated sequence of tokens, including the initial prefix and any additional tokens generated by the function.
    """

Citation

Please cite our paper if you find our work valuable.

@article{ouroboros,
  title={Ouroboros: Speculative Decoding with Large Model Enhanced Drafting},
  author={Weilin Zhao, Yuxiang Huang, Xu Han, Chaojun Xiao, Zhiyuan Liu, Maosong Sun},
  journal={arXiv preprint arXiv:2402.13720},
  url={https://arxiv.org/abs/2402.13720}
  year={2024}
}