metadata

language:
  - en
tags:
  - pytorch
  - causal-lm
  - dcformer
  - dcmha
license: mit

DCPythia-6.9B is a pretrained language model on the Pile with 300B tokens. With comparison of Pythia-6.9B, we validate the scaling performance of Dynamically Composable Multi-Head Attention(DCMHA), a parameter and computation efficient attention architecture that tackles the shortcomings of MHA and increases the expressive power of the model by dynamically composing attention heads. Please see downstrem evaluations and more details in the paper(Improving Transformers with Dynamically Composable Multi-Head Attention). In addition, we open-source Jax training code on (Github).

We recommend compiled version of DCPythia with torch.compile for inference acceleration. Please refer to Generation section for compile implementation.

Usage

Env

You need to upgrade transformers to avoid (loading problems).

pip install transformers>=4.40.2

Generation

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

import os
os.environ['TOKENIZERS_PARALLELISM'] = 'false'

tokenizer = AutoTokenizer.from_pretrained("Caiyun-AI/DCPythia-6.9B")
model = AutoModelForCausalLM.from_pretrained("Caiyun-AI/DCPythia-6.9B", trust_remote_code=True)

device = torch.device('cuda')
MAX_BATCH_SIZE = 1
MAX_SEQ_LENGTH = 2048
NUM_TOKENS_TO_GENERATE = 100
COMPILE = True

_ = model.to(device=device,dtype=torch.float16)
with torch.device(device):
    model.setup_caches(max_batch_size=MAX_BATCH_SIZE, max_seq_length=MAX_SEQ_LENGTH, set_kv_cache=True)

def decode_one_token(model, cur_token, input_pos):
    logits = model(cur_token, input_pos=input_pos, return_tensor=True)
    new_token = torch.argmax(logits[:, -1], dim=-1)[:,None]
    return new_token

prompt = "Beijing is the capital of China. London is the capital of"
input_ids = tokenizer.encode(prompt, return_tensors='pt')

compiled_decode_one_token = torch.compile(decode_one_token,mode="reduce-overhead", fullgraph=True) if COMPILE else None

with torch.no_grad():
    generated_ids = model.generate(input_ids.to(device),num_tokens_to_generate=NUM_TOKENS_TO_GENERATE, compiled_decode_one_token=compiled_decode_one_token)
    text = tokenizer.decode(generated_ids[0])
    print('generated text:', text)