Feedback Transformer - Pytorch
Simple implementation of Feedback Transformer in Pytorch. They improve on Transformer-XL by having each token have access to the representations of all previous layers through time. This is achieved by aggregating the outputs of all layers into a shared memory, which each token across layers can attend to at each time step.
The main drawback is longer training time, due to its non-parallel nature. But I thought I'd build it to further exploration and research into this line of work.
I also took the liberty to add some various enhancements, including pre-normalization, GLU gated feedforwards, as well as simplified T5 relative positional embeddings.
Install
$ pip install feedback-transformer-pytorch
Usage
import torch
from feedback_transformer_pytorch import FeedbackTransformer
model = FeedbackTransformer(
num_tokens = 20000, # number of tokens
dim = 512, # dimension
depth = 6, # depth
seq_len = 2, # the sequence length of each segment or window
mem_len = 256, # length of the memory buffer
dim_head = 64, # dimension of each head
heads = 8, # number of heads
attn_dropout = 0.1, # attention dropout
ff_dropout = 0.1 # feedforward dropout
).cuda()
x = torch.randint(0, 20000, (2, 64)).cuda()
model(x) # (2, 64, 20000)
If you would like to have fine control over the memory (when to detach, etc), you can do it with some extra keyword arguments on .forward
import torch
from feedback_transformer_pytorch import FeedbackTransformer
model = FeedbackTransformer(
num_tokens = 20000,
dim = 512,
depth = 6,
seq_len = 32,
mem_len = 256
).cuda()
x1 = torch.randint(0, 20000, (2, 32)).cuda()
x2 = torch.randint(0, 20000, (2, 32)).cuda()
x3 = torch.randint(0, 20000, (2, 32)).cuda()
out1, mem1 = model(x1, return_memory = True)
out2, mem2 = model(x2, memory = mem1, return_memory = True)
out3, mem3 = model(x3, memory = mem2, return_memory = True) # (2, 32, 20000)
Citations
@misc{fan2021addressing,
title = {Addressing Some Limitations of Transformers with Feedback Memory},
author = {Angela Fan and Thibaut Lavril and Edouard Grave and Armand Joulin and Sainbayar Sukhbaatar},
year = {2021},
eprint = {2002.09402},
archivePrefix = {arXiv},
primaryClass = {cs.LG}
}
53 Dec 02, 2022
6 Dec 10, 2021
94 Dec 18, 2022
6 Sep 22, 2022
1 Nov 22, 2021
293 Jan 06, 2023
9 Jun 16, 2022
0 Feb 06, 2022
118 Dec 29, 2022
29 Nov 08, 2022
397 Dec 27, 2022
60 Oct 21, 2022
11 Nov 27, 2022
19 Dec 12, 2022
12 Mar 29, 2022
67 Dec 20, 2022
25.7k Jan 09, 2023
0 Feb 26, 2022
1 Feb 08, 2022
57 Nov 24, 2022