routing transformer
1.6.1
Eine voll funktionsfähige Implementierung von Routing Transformer. Das Papier schlägt vor, k-means zu verwenden, um ähnliche Abfragen/Schlüssel zur Aufmerksamkeit an denselben Cluster weiterzuleiten.
131.000 Token
$ pip install routing_transformerEin einfaches Sprachmodell
import torch
from routing_transformer import RoutingTransformerLM
model = RoutingTransformerLM (
num_tokens = 20000 ,
dim = 512 ,
heads = 8 ,
depth = 12 ,
max_seq_len = 8192 ,
causal = True , # auto-regressive or not
emb_dim = 128 , # embedding factorization, from Albert
weight_tie = False , # weight tie layers, from Albert
tie_embedding = False , # multiply final embeddings with token weights for logits
dim_head = 64 , # be able to fix the dimension of each head, making it independent of the embedding dimension and the number of heads
attn_dropout = 0.1 , # dropout after attention
attn_layer_dropout = 0. , # dropout after self attention layer
ff_dropout = 0.1 , # feedforward dropout
layer_dropout = 0. , # layer dropout
window_size = 128 , # target window size of each cluster
n_local_attn_heads = 4 , # number of local attention heads
reversible = True , # reversible networks for memory savings, from Reformer paper
ff_chunks = 10 , # feed forward chunking, from Reformer paper
ff_glu = True , # use GLU variant in feedforward
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
moe_layers = ( 3 , 6 ), # specify which layers to use mixture of experts
moe_num_experts = 4 , # number of experts in the mixture of experts layer, defaults to 4. increase for adding more parameters to model
moe_loss_coef = 1e-2 , # the weight for the auxiliary loss in mixture of experts to keep expert usage balanced
num_mem_kv = 8 , # number of memory key/values to append to each cluster of each head, from the 'All-Attention' paper. defaults to 1 in the causal case for unshared QK to work
use_scale_norm = False , # use scale norm, simplified normalization from 'Transformers without Tears' paper
use_rezero = False , # use Rezero with no normalization
shift_tokens = True # shift tokens by one along sequence dimension, for a slight improvement in convergence
). cuda ()
x = torch . randint ( 0 , 20000 , ( 1 , 8192 )). long (). cuda ()
input_mask = torch . ones_like ( x ). bool (). cuda ()
y , aux_loss = model ( x , input_mask = input_mask ) # (1, 8192, 20000)
aux_loss . backward () # add auxiliary loss to main loss before backpropEin einfacher Transformator
import torch
from routing_transformer import RoutingTransformer
model = RoutingTransformer (
dim = 512 ,
heads = 8 ,
depth = 12 ,
max_seq_len = 8192 ,
window_size = 128 ,
n_local_attn_heads = 4
). cuda ()
x = torch . randn ( 1 , 8192 , 512 ). cuda ()
input_mask = torch . ones ( 1 , 8192 ). bool (). cuda ()
y , aux_loss = model ( x , input_mask = input_mask ) # (1, 8192, 512)
aux_loss . backward () # add auxiliary loss to main loss before backprop Um einen vollständigen Encoder oder Decoder zu verwenden, importieren Sie einfach die RoutingTransformerEncDec -Klasse. Mit Ausnahme des Schlüsselworts dim wird allen anderen Schlüsselwörtern entweder enc_ oder dec_ für die Encoder- bzw. Decoder RoutingTransformerLM Klasse vorangestellt.
import torch
from routing_transformer import RoutingTransformerEncDec
model = RoutingTransformerEncDec (
dim = 512 ,
enc_num_tokens = 20000 ,
enc_depth = 4 ,
enc_heads = 8 ,
enc_max_seq_len = 4096 ,
enc_window_size = 128 ,
dec_num_tokens = 20000 ,
dec_depth = 4 ,
dec_heads = 8 ,
dec_max_seq_len = 4096 ,
dec_window_size = 128 ,
dec_reversible = True
). cuda ()
src = torch . randint ( 0 , 20000 , ( 1 , 4096 )). cuda ()
tgt = torch . randint ( 0 , 20000 , ( 1 , 4096 )). cuda ()
src_mask = torch . ones_like ( src ). bool (). cuda ()
tgt_mask = torch . ones_like ( tgt ). bool (). cuda ()
loss , aux_loss = model ( src , tgt , enc_input_mask = src_mask , dec_input_mask = tgt_mask , return_loss = True , randomly_truncate_sequence = True )
loss . backward ()
aux_loss . backward ()
# do your training, then to sample up to 2048 tokens based on the source sequence
src = torch . randint ( 0 , 20000 , ( 1 , 4096 )). cuda ()
start_tokens = torch . ones ( 1 , 1 ). long (). cuda () # assume starting token is 1
sample = model . generate ( src , start_tokens , seq_len = 2048 , eos_token = 2 ) # (1, <= 2048, 20000) Um die Vorteile der Verwendung von PKM zu erkennen, muss die Lernrate der Werte höher eingestellt werden als die der übrigen Parameter. (Empfohlen 1e-2 )
Sie können den Anweisungen hier folgen, um es richtig einzustellen: https://github.com/lucidrains/product-key-memory#learning-rates
kmeans_ema_decay = {defaults to 0.999}Dies ist der exponentielle Abfall des gleitenden Durchschnitts zur Aktualisierung der k-Mittelwerte. Je niedriger dieser Wert ist, desto schneller passen sich die Mittel an, allerdings auf Kosten der Stabilität.
commitment_factor = {defaults to 1e-4}Das Gewicht des Hilfsverlusts, der Token dazu ermutigt, sich den für sie ausgewählten k-Mittelwertschwerpunkten anzunähern (festzuschreiben).
Mit den folgenden Anweisungen können Sie die kmeans manuell aktualisieren. Standardmäßig werden die kmeans bei jedem Rückwärtsdurchlauf automatisch aktualisiert.
import torch
from routing_transformer import RoutingTransformerLM , AutoregressiveWrapper
model = RoutingTransformerLM (
num_tokens = 20000 ,
dim = 1024 ,
heads = 8 ,
depth = 6 ,
window_size = 256 ,
max_seq_len = 8192 ,
causal = True ,
_register_kmeans_update = False # set to False to disable auto-updating
)
model = AutoregressiveWrapper ( model )
x = torch . randint ( 0 , 20000 , ( 1 , 8192 ))
loss = model ( x , return_loss = True )
loss . backward ()
# update kmeans with this call
model . update_kmeans ()Diese Architektur hat Probleme bei der Verallgemeinerung auf kürzere Sequenzlängen, wenn Token von 1 -> maximale Sequenzlänge dekodiert werden. Die einfachste und sicherste Lösung besteht darin, die Sequenz während des Trainings zufällig abzuschneiden. Dies hilft dem Netzwerk und den kmeans, sich auf eine variable Anzahl von Token zu verallgemeinern, allerdings auf Kosten einer längeren Schulung.
Wenn Sie das Netzwerk zu Beginn mit der vollen Sequenzlänge vorbereiten, tritt dieses Problem nicht auf und Sie können diesen Trainingsvorgang überspringen.
import torch
from routing_transformer import RoutingTransformerLM , AutoregressiveWrapper
model = RoutingTransformerLM (
num_tokens = 20000 ,
dim = 1024 ,
heads = 8 ,
depth = 12 ,
window_size = 256 ,
max_seq_len = 8192 ,
causal = True
)
model = AutoregressiveWrapper ( model )
x = torch . randint ( 0 , 20000 , ( 1 , 8192 ))
loss = model ( x , return_loss = True , randomly_truncate_sequence = True ) # (1, 8192, 20000) Besonderer Dank geht an Aran Komatsuzaki für das Bootstrapping der ersten Implementierung in Pytorch, die zu dieser Bibliothek führte.
@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://arxiv.org/pdf/2003.05997.pdf }
} @misc { shazeer2020glu ,
title = { GLU Variants Improve Transformer } ,
author = { Noam Shazeer } ,
year = { 2020 } ,
url = { https://arxiv.org/abs/2002.05202 }
} @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 }
} @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 }
} @misc { lan2019albert ,
title = { ALBERT: A Lite BERT for Self-supervised Learning of Language Representations } ,
author = { Zhenzhong Lan and Mingda Chen and Sebastian Goodman and Kevin Gimpel and Piyush Sharma and Radu Soricut } ,
year = { 2019 } ,
url = { https://arxiv.org/abs/1909.11942 }
} @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 }
} @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 }
} @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 }
} @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 } ,
} @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 { 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 }
} @software { peng_bo_2021_5196578 ,
author = { PENG Bo } ,
title = { BlinkDL/RWKV-LM: 0.01 } ,
month = { aug } ,
year = { 2021 } ,
publisher = { Zenodo } ,
version = { 0.01 } ,
doi = { 10.5281/zenodo.5196578 } ,
url = { https://doi.org/10.5281/zenodo.5196578 }
}
