issues
search
awslabs
/
sockeye
Sequence-to-sequence framework with a focus on Neural Machine Translation based on PyTorch
https://awslabs.github.io/sockeye/
Apache License 2.0
1.21k
stars
323
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Add --tf32 device flag for transparent float32 acceleration
#1066
graehl
closed
2 years ago
5
BrokenPipeError: [Errno 32]
#1065
whher
closed
2 years ago
6
translator: send skip_nvs and nvs_thresh from args to ctor
#1064
graehl
closed
2 years ago
5
Train NVS using Sockeye
#1063
HossamAmer12
closed
2 years ago
11
Add kNN-MT
#1062
shuoyangd
closed
1 year ago
12
Data streaming in Sockeye 3
#1061
tomsbergmanis
closed
2 years ago
2
Implement the Scones paper
#1060
jvamvas
closed
1 year ago
1
Add Support for DeepSpeed ZeRO Stage 1
#1059
mjdenkowski
closed
2 years ago
2
Update README.md with some recent publications
#1058
fhieber
closed
2 years ago
0
[Review #1051 First] Refactoring Ahead of Adding DeepSpeed Support
#1057
mjdenkowski
closed
2 years ago
2
Update website with latest arxiv paper
#1056
fhieber
closed
2 years ago
0
Out of memory exception after releasing GPUs in epoch 0
#1055
ejkitchen
closed
2 years ago
2
Guided alignments ala Garg et al.
#1054
tomsbergmanis
closed
1 year ago
7
About the Arxiv benchmark and the paper
#1053
vince62s
closed
2 years ago
13
Add Sockeye 3 paper to citations and BibTeX file
#1052
mjdenkowski
closed
2 years ago
0
Option to clamp layer outputs to min/max finite values for their dtypes
#1051
mjdenkowski
closed
2 years ago
0
Support offline model quantization
#1050
mjdenkowski
closed
2 years ago
0
How to plot attention matrix
#1049
etexaco123
closed
2 years ago
1
Nbest reranking options using isometric translation criteria.
#1048
surafelml
closed
2 years ago
9
Fix data type of valid_length
#1047
hoangcuong2011
closed
2 years ago
0
Neural vocabulary selection.
#1046
tdomhan
closed
2 years ago
0
Option to not reload the best training checkpoint when reducing the learning rate
#1045
mjdenkowski
closed
2 years ago
0
Fix scoring with batches of size 1 by avoid squeezing the batch dim
#1044
fhieber
closed
2 years ago
0
Do not create incorrect checkpoints when resuming training with a fully trained model
#1043
mjdenkowski
closed
2 years ago
0
Ignore false-positive missing keys for traced modules when loading model parameters
#1042
mjdenkowski
closed
2 years ago
0
Update Readme.md with recent publications
#1041
fhieber
closed
2 years ago
0
Model parameter keys are prefixed with "traced"
#1040
bricksdont
closed
2 years ago
10
Using SLURM, 2 nodes each with 4 GPUs [problem]
#1039
SamuelLarkin
closed
2 years ago
13
ValueError: Unsupported nproc_per_node value: 4(x2)
#1038
SamuelLarkin
closed
2 years ago
1
Add comment about "traced" naming convention
#1037
mjdenkowski
closed
2 years ago
0
Do not save parameters for traced modules.
#1036
mjdenkowski
closed
2 years ago
0
Optimized creation of repeated_state by using expand instead of repeat_interleave.
#1035
hoangcuong2011
closed
2 years ago
3
I have challenges during sockeye training
#1034
etexaco123
closed
2 years ago
1
Clarify use of Translator.batch_size in code
#1033
fhieber
closed
2 years ago
0
Also trace SockeyeModel components when `inference_only == False` (includes CheckpointDecoder)
#1032
mjdenkowski
closed
2 years ago
0
Moved offsetting of topk scores out of the (traced) TopK module.
#1031
fhieber
closed
2 years ago
0
I need some help with sockeye 3.1
#1030
etexaco123
closed
2 years ago
1
Allow PyTorch 1.11 in requirements
#1029
fhieber
closed
2 years ago
0
Code formatting & cleanup for beam_search.py and inference.py
#1028
fhieber
closed
2 years ago
0
Vectors/Images/Video to Text Translation
#1027
AmitMY
closed
1 year ago
4
SockeyeError: Model(s) require 1 factors, but 2 given (through --input and --input-factors).
#1026
RamoramaInteractive
closed
2 years ago
1
Support target prefix with JSON format
#1025
hoangcuong2011
closed
2 years ago
5
support JSON source prefix
#1024
hoangcuong2011
closed
2 years ago
4
Faster prepare_source_length_mask using expand
#1023
fhieber
closed
2 years ago
0
Integration with huggingface
#1022
bricksdont
closed
1 year ago
8
Allow torch==1.10.2, rename unit test file
#1021
fhieber
closed
2 years ago
0
Fixed GPU-based scoring by copying to cpu tensor first before converting to numpy.
#1020
fhieber
closed
2 years ago
1
Source-Factoring Translation not as expected
#1019
RamoramaInteractive
closed
2 years ago
3
Added sacrebleu's TER metric to checkpoint decoder metrics.
#1018
fhieber
closed
2 years ago
0
Mismatch in arguments for training continuation. Differing arguments: validation_target, validation_source
#1017
RamoramaInteractive
closed
2 years ago
2
Previous
Next