reformer pytorch
1.4.4
Esta es una implementación de Pytorch de reformador https://openreview.net/pdf?id=rkgnkhtvb
Incluye atención LSH, red reversible y fragmentación. Se ha validado con una tarea auto-regresiva (enwik8).
Tokens de 32k
Tokens 81k con media precisión
$ pip install reformer_pytorchUn simple modelo de idioma reformador
# should fit in ~ 5gb - 8k tokens
import torch
from reformer_pytorch import ReformerLM
model = ReformerLM (
num_tokens = 20000 ,
dim = 1024 ,
depth = 12 ,
max_seq_len = 8192 ,
heads = 8 ,
lsh_dropout = 0.1 ,
ff_dropout = 0.1 ,
post_attn_dropout = 0.1 ,
layer_dropout = 0.1 , # layer dropout from 'Reducing Transformer Depth on Demand' paper
causal = True , # auto-regressive or not
bucket_size = 64 , # average size of qk per bucket, 64 was recommended in paper
n_hashes = 4 , # 4 is permissible per author, 8 is the best but slower
emb_dim = 128 , # embedding factorization for further memory savings
dim_head = 64 , # be able to fix the dimension of each head, making it independent of the embedding dimension and the number of heads
ff_chunks = 200 , # number of chunks for feedforward layer, make higher if there are memory issues
attn_chunks = 8 , # process lsh attention in chunks, only way for memory to fit when scaling to 16k tokens
num_mem_kv = 128 , # persistent learned memory key values, from all-attention paper
full_attn_thres = 1024 , # use full attention if context length is less than set value
reverse_thres = 1024 , # turn off reversibility for 2x speed for sequence lengths shorter or equal to the designated value
use_scale_norm = False , # use scale norm from 'Transformers without tears' paper
use_rezero = False , # remove normalization and use rezero from 'ReZero is All You Need'
one_value_head = False , # use one set of values for all heads from 'One Write-Head Is All You Need'
weight_tie = False , # tie parameters of each layer for no memory per additional depth
weight_tie_embedding = False , # use token embedding for projection of output, some papers report better results
n_local_attn_heads = 2 , # many papers suggest mixing local attention heads aids specialization and improves on certain tasks
pkm_layers = ( 4 , 7 ), # specify layers to use product key memory. paper shows 1 or 2 modules near the middle of the transformer is best
pkm_num_keys = 128 , # defaults to 128, but can be increased to 256 or 512 as memory allows
use_full_attn = False # only turn on this flag to override and turn on full attention for all sequence lengths. for comparison with LSH to show that it is working
). cuda ()
x = torch . randint ( 0 , 20000 , ( 1 , 8192 )). long (). cuda ()
y = model ( x ) # (1, 8192, 20000)El reformador (solo una pila de atención reversible LSH)
# should fit in ~ 5gb - 8k embeddings
import torch
from reformer_pytorch import Reformer
model = Reformer (
dim = 512 ,
depth = 12 ,
heads = 8 ,
lsh_dropout = 0.1 ,
causal = True
). cuda ()
x = torch . randn ( 1 , 8192 , 512 ). cuda ()
y = model ( x ) # (1, 8192, 512)Atención propia con LSH
import torch
from reformer_pytorch import LSHSelfAttention
attn = LSHSelfAttention (
dim = 128 ,
heads = 8 ,
bucket_size = 64 ,
n_hashes = 8 ,
causal = False
)
x = torch . randn ( 10 , 1024 , 128 )
y = attn ( x ) # (10, 1024, 128)Atención LSH (hashing sensible a la localidad)
import torch
from reformer_pytorch import LSHAttention
attn = LSHAttention (
bucket_size = 64 ,
n_hashes = 16 ,
causal = True
)
qk = torch . randn ( 10 , 1024 , 128 )
v = torch . randn ( 10 , 1024 , 128 )
out , attn , buckets = attn ( qk , v ) # (10, 1024, 128)
# attn contains the unsorted attention weights, provided return_attn is set to True (costly otherwise)
# buckets will contain the bucket number (post-argmax) of each token of each batch Este repositorio admite máscaras en la secuencia de entrada input_mask (bx i_seq) , la secuencia de contexto context_mask (bx c_seq) , así como la matriz de atención completa rara vez usada en input_attn_mask (bx i_seq x i_seq) , todo se hace compatible con la atención LSH. Las máscaras están hechas de booleanos donde False denotas enmascarar antes del Softmax.
La máscara triangular causal se cuida por usted si establece causal = True .
import torch
from reformer_pytorch import ReformerLM
CONTEXT_LEN = 512
SEQ_LEN = 8192
model = ReformerLM (
num_tokens = 20000 ,
dim = 1024 ,
depth = 1 ,
max_seq_len = SEQ_LEN ,
ff_chunks = 8 ,
causal = True
)
c = torch . randn ( 1 , CONTEXT_LEN , 1024 )
x = torch . randint ( 0 , 20000 , ( 1 , SEQ_LEN )). long ()
i_mask = torch . ones ( 1 , SEQ_LEN ). bool ()
c_mask = torch . ones ( 1 , CONTEXT_LEN ). bool ()
y = model ( x , keys = c , input_mask = i_mask , context_mask = c_mask )
# masking done correctly in LSH attention La incrustación posicional predeterminada utiliza incrustaciones rotativas.
Sin embargo, Aran me ha informado que el equipo de reformador usó incrustaciones de posición axial con excelentes resultados en secuencias más largas.
Puede activar la incrustación posicional axial y ajustar la forma y la dimensión de los incrustaciones axiales siguiendo las instrucciones a continuación.
import torch
from reformer_pytorch import ReformerLM
model = ReformerLM (
num_tokens = 20000 ,
dim = 1024 ,
depth = 12 ,
max_seq_len = 8192 ,
ff_chunks = 8 ,
attn_chunks = 2 ,
causal = True ,
axial_position_emb = True , # set this to True
axial_position_shape = ( 128 , 64 ), # the shape must multiply up to the max_seq_len (128 x 64 = 8192)
)
x = torch . randint ( 0 , 20000 , ( 1 , 8192 )). long ()
y = model ( x ) # (1, 8192, 20000) Si prefiere usar incrustaciones posicionales absolutas, puede encenderlo con absolute_position_emb = True indicador en la inicialización.
Desde la versión 0.17.0 , y algunas correcciones a la red reversible, ¡el reformador Pytorch es compatible con la velocidad profunda de Microsoft! Si tiene varias GPU locales, puede seguir las instrucciones / ejemplo aquí.
Una secuencia de reformador completa → secuencia, digamos la traducción
import torch
from reformer_pytorch import ReformerLM
DE_SEQ_LEN = 4096
EN_SEQ_LEN = 4096
encoder = ReformerLM (
num_tokens = 20000 ,
emb_dim = 128 ,
dim = 1024 ,
depth = 12 ,
heads = 8 ,
max_seq_len = DE_SEQ_LEN ,
fixed_position_emb = True ,
return_embeddings = True # return output of last attention layer
). cuda ()
decoder = ReformerLM (
num_tokens = 20000 ,
emb_dim = 128 ,
dim = 1024 ,
depth = 12 ,
heads = 8 ,
max_seq_len = EN_SEQ_LEN ,
fixed_position_emb = True ,
causal = True
). cuda ()
x = torch . randint ( 0 , 20000 , ( 1 , DE_SEQ_LEN )). long (). cuda ()
yi = torch . randint ( 0 , 20000 , ( 1 , EN_SEQ_LEN )). long (). cuda ()
enc_keys = encoder ( x ) # (1, 4096, 1024)
yo = decoder ( yi , keys = enc_keys ) # (1, 4096, 20000)Una imagen de reformador completa → subtítulos
import torch
from torch . nn import Sequential
from torchvision import models
from reformer_pytorch import Reformer , ReformerLM
resnet = models . resnet50 ( pretrained = True )
resnet = Sequential ( * list ( resnet . children ())[: - 4 ])
SEQ_LEN = 4096
encoder = Reformer (
dim = 512 ,
depth = 6 ,
heads = 8 ,
max_seq_len = 4096
)
decoder = ReformerLM (
num_tokens = 20000 ,
dim = 512 ,
depth = 6 ,
heads = 8 ,
max_seq_len = SEQ_LEN ,
causal = True
)
x = torch . randn ( 1 , 3 , 512 , 512 )
yi = torch . randint ( 0 , 20000 , ( 1 , SEQ_LEN )). long ()
visual_emb = resnet ( x )
b , c , h , w = visual_emb . shape
visual_emb = visual_emb . view ( 1 , c , h * w ). transpose ( 1 , 2 ) # nchw to nte
enc_keys = encoder ( visual_emb )
yo = decoder ( yi , keys = enc_keys ) # (1, 4096, 20000) Hay un error en las versiones < 0.21.0 . Actualice al menos a la versión especificada para el codificador de trabajo / reformador de decodificadores.
Por demanda popular, he codificado un envoltorio que elimina gran parte del trabajo manual al escribir una arquitectura de codificador / decodificador de reformador genérico. Para usar, importaría la clase ReformerEncDec . Los argumentos de palabras clave del codificador se pasarían con un prefijo enc_ y los argumentos de palabras clave del decodificador con dec_ . La dimensión del modelo ( dim ) debe estar libre de prefijo y se compartirá entre el codificador y el decodificador. El marco también se encargará de pasar la máscara de entrada del codificador a la máscara de contexto del decodificador, a menos que se anule explícitamente.
import torch
from reformer_pytorch import ReformerEncDec
DE_SEQ_LEN = 4096
EN_SEQ_LEN = 4096
enc_dec = ReformerEncDec (
dim = 512 ,
enc_num_tokens = 20000 ,
enc_depth = 6 ,
enc_max_seq_len = DE_SEQ_LEN ,
dec_num_tokens = 20000 ,
dec_depth = 6 ,
dec_max_seq_len = EN_SEQ_LEN
). cuda ()
train_seq_in = torch . randint ( 0 , 20000 , ( 1 , DE_SEQ_LEN )). long (). cuda ()
train_seq_out = torch . randint ( 0 , 20000 , ( 1 , EN_SEQ_LEN )). long (). cuda ()
input_mask = torch . ones ( 1 , DE_SEQ_LEN ). bool (). cuda ()
loss = enc_dec ( train_seq_in , train_seq_out , return_loss = True , enc_input_mask = input_mask )
loss . backward ()
# learn
# evaluate with the following
eval_seq_in = torch . randint ( 0 , 20000 , ( 1 , DE_SEQ_LEN )). long (). cuda ()
eval_seq_out_start = torch . tensor ([[ 0. ]]). long (). cuda () # assume 0 is id of start token
samples = enc_dec . generate ( eval_seq_in , eval_seq_out_start , seq_len = EN_SEQ_LEN , eos_token = 1 ) # assume 1 is id of stop token
print ( samples . shape ) # (1, <= 1024) decode the tokens Para ver los beneficios de usar PKM, la tasa de aprendizaje de los valores debe establecerse más alta que el resto de los parámetros. (Recomendado para ser 1e-2 )
Puede seguir las instrucciones aquí para configurarlo correctamente https://github.com/lucidrains/product-key-memory#learning-rates
Por defecto, la función de activación es GELU . Si desea una función de activación alternativa, puede pasar en la clase a la palabra clave ff_activation .
import torch
from reformer_pytorch import ReformerLM
from torch import nn
model = ReformerLM (
num_tokens = 20000 ,
dim = 512 ,
depth = 6 ,
max_seq_len = 8192 ,
ff_chunks = 8 ,
ff_dropout = 0.1 ,
ff_mult = 6 ,
ff_activation = nn . LeakyReLU ,
ff_glu = True # use GLU in feedforward, from paper 'GLU Variants Improve Transformer'
)
x = torch . randint ( 0 , 20000 , ( 1 , 8192 )). long ()
y = model ( x ) # (1, 8192, 20000) Para acceder a los pesos de atención y la distribución de deseos, simplemente envuelva el modelo instanciado con la clase de envoltura Recorder .
import torch
from reformer_pytorch import Reformer , Recorder
model = Reformer (
dim = 512 ,
depth = 12 ,
max_seq_len = 8192 ,
heads = 8 ,
lsh_dropout = 0.1 ,
causal = True
). cuda ()
model = Recorder ( model )
x = torch . randn ( 1 , 8192 , 512 ). cuda ()
y = model ( x )
model . recordings [ 0 ] # a list of attention weights and buckets for the first forward pass
model . turn_off () # stop recording
model . turn_on () # start recording
model . clear () # clear the recordings
model = model . eject () # recover the original model and remove all listeners Reformer viene con un ligero inconveniente de que la secuencia debe ser perfectamente divisible por el tamaño del cubo * 2. He proporcionado una pequeña herramienta a ayuda de ayuda que puede ayudarlo a superar automáticamente la longitud de la secuencia a la siguiente mejor múltiplo.
import torch
from reformer_pytorch import ReformerLM , Autopadder
model = ReformerLM (
num_tokens = 20000 ,
dim = 1024 ,
depth = 12 ,
max_seq_len = 8192 ,
heads = 8 ,
lsh_dropout = 0.1 ,
causal = True ,
bucket_size = 63 , # odd bucket size
num_mem_kv = 77 # odd memory key length
). cuda ()
model = Autopadder ( model )
SEQ_LEN = 7777 # odd sequence length
keys = torch . randn ( 1 , 137 , 1024 ) # odd keys length
x = torch . randint ( 0 , 20000 , ( 1 , SEQ_LEN )). long (). cuda ()
y = model ( x , keys = keys ) # (1, 7777, 20000) Muchos usuarios solo están interesados en un modelo de idioma auto-regresivo (como GPT-2). Aquí hay un envoltorio de entrenamiento para facilitar el entrenamiento y evaluar en secuencias de tokens codificados arbitrariamente. Tendrá que cuidar la codificación y la decodificación.
import torch
from torch import randint
from reformer_pytorch import ReformerLM
from reformer_pytorch . generative_tools import TrainingWrapper
model = ReformerLM (
num_tokens = 20000 ,
dim = 1024 ,
depth = 12 ,
max_seq_len = 4096 ,
lsh_dropout = 0.1 ,
causal = True ,
full_attn_thres = 1024
)
# 0 is used for padding and no loss to be calculated on it
model = TrainingWrapper ( model , ignore_index = 0 , pad_value = 0 )
# the wrapper can handle evenly packed sequences
x_train = randint ( 0 , 20000 , ( 3 , 357 ))
# or if you have a list of uneven sequences, it will be padded for you
x_train = [
randint ( 0 , 20000 , ( 120 ,)),
randint ( 0 , 20000 , ( 253 ,)),
randint ( 0 , 20000 , ( 846 ,))
]
# when training, set return_loss equal to True
model . train ()
loss = model ( x_train , return_loss = True )
loss . backward ()
# when evaluating, just use the generate function, which will default to top_k sampling with temperature of 1.
initial = torch . tensor ([[ 0 ]]). long () # assume 0 is start token
sample = model . generate ( initial , 100 , temperature = 1. , filter_thres = 0.9 , eos_token = 1 ) # assume end token is 1, or omit and it will sample up to 100
print ( sample . shape ) # (1, <=100) token ids Andrea ha descubierto que usar el nivel de optimización de O2 cuando el entrenamiento con precisión mixta puede conducir a la inestabilidad. Utilice O1 en su lugar, que se puede configurar con amp_level en Pytorch Lightning u opt_level en la biblioteca Apex de NVIDIA.
@inproceedings { kitaev2020reformer ,
title = { Reformer: The Efficient Transformer } ,
author = { Nikita Kitaev and Lukasz Kaiser and Anselm Levskaya } ,
booktitle = { International Conference on Learning Representations } ,
year = { 2020 } ,
url = { https://openreview.net/forum?id=rkgNKkHtvB }
} @article { DBLP:journals/corr/abs-1907-01470 ,
author = { Sainbayar Sukhbaatar and
Edouard Grave and
Guillaume Lample and
Herv{'{e}} J{'{e}}gou and
Armand Joulin } ,
title = { Augmenting Self-attention with Persistent Memory } ,
journal = { CoRR } ,
volume = { abs/1907.01470 } ,
year = { 2019 } ,
url = { http://arxiv.org/abs/1907.01470 }
} @article { 1910.05895 ,
author = { Toan Q. Nguyen and Julian Salazar } ,
title = { Transformers without Tears: Improving the Normalization of Self-Attention } ,
year = { 2019 } ,
eprint = { arXiv:1910.05895 } ,
doi = { 10.5281/zenodo.3525484 } ,
} @inproceedings { fan2020reducing ,
title = { Reducing Transformer Depth on Demand with Structured Dropout } ,
author = { Angela Fan and Edouard Grave and Armand Joulin } ,
booktitle = { International Conference on Learning Representations } ,
year = { 2020 } ,
url = { https://openreview.net/forum?id=SylO2yStDr }
} @article { Shazeer2019FastTD ,
title = { Fast Transformer Decoding: One Write-Head is All You Need } ,
author = { Noam Shazeer } ,
journal = { ArXiv } ,
year = { 2019 } ,
volume = { abs/1911.02150 }
} @misc { shazeer2020glu ,
title = { GLU Variants Improve Transformer } ,
author = { Noam Shazeer } ,
year = { 2020 } ,
url = { https://arxiv.org/abs/2002.05202 }
} @misc { roy*2020efficient ,
title = { Efficient Content-Based Sparse Attention with Routing Transformers } ,
author = { Aurko Roy* and Mohammad Taghi Saffar* and David Grangier and Ashish Vaswani } ,
year = { 2020 } ,
url = { https://openreview.net/forum?id=B1gjs6EtDr }
} @misc { bachlechner2020rezero ,
title = { ReZero is All You Need: Fast Convergence at Large Depth } ,
author = { Thomas Bachlechner and Bodhisattwa Prasad Majumder and Huanru Henry Mao and Garrison W. Cottrell and Julian McAuley } ,
year = { 2020 } ,
url = { https://arxiv.org/abs/2003.04887 }
} @misc { lample2019large ,
title = { Large Memory Layers with Product Keys } ,
author = { Guillaume Lample and Alexandre Sablayrolles and Marc'Aurelio Ranzato and Ludovic Denoyer and Hervé Jégou } ,
year = { 2019 } ,
eprint = { 1907.05242 } ,
archivePrefix = { arXiv }
} @misc { bhojanapalli2020lowrank ,
title = { Low-Rank Bottleneck in Multi-head Attention Models } ,
author = { Srinadh Bhojanapalli and Chulhee Yun and Ankit Singh Rawat and Sashank J. Reddi and Sanjiv Kumar } ,
year = { 2020 } ,
eprint = { 2002.07028 }
} @misc { dong2021attention ,
title = { Attention is Not All You Need: Pure Attention Loses Rank Doubly Exponentially with Depth } ,
author = { Yihe Dong and Jean-Baptiste Cordonnier and Andreas Loukas } ,
year = { 2021 } ,
eprint = { 2103.03404 }
} @misc { su2021roformer ,
title = { RoFormer: Enhanced Transformer with Rotary Position Embedding } ,
author = { Jianlin Su and Yu Lu and Shengfeng Pan and Bo Wen and Yunfeng Liu } ,
year = { 2021 } ,
eprint = { 2104.09864 } ,
archivePrefix = { arXiv } ,
primaryClass = { cs.CL }
} @misc { vaswani2017attention ,
title = { Attention Is All You Need } ,
author = { Ashish Vaswani and Noam Shazeer and Niki Parmar and Jakob Uszkoreit and Llion Jones and Aidan N. Gomez and Lukasz Kaiser and Illia Polosukhin } ,
year = { 2017 } ,
eprint = { 1706.03762 } ,
archivePrefix = { arXiv } ,
primaryClass = { cs.CL }
}♥
