Ecosyste.ms: Issues
An open API service for providing issue and pull request metadata for open source projects.
GitHub / awslabs/sockeye issues and pull requests
#1116 - Retire DeepSpeed mode due to security issue with supported version
Pull Request -
State: closed - Opened by mjdenkowski about 1 month ago
#1115 - Bump deepspeed from 0.6.5 to 0.15.1 in /requirements
Pull Request -
State: closed - Opened by dependabot[bot] about 1 month ago
- 1 comment
Labels: dependencies
#1114 - Add maintenance mode notice to README.
Pull Request -
State: closed - Opened by mjdenkowski 6 months ago
- 3 comments
#1113 - Wrong parameter when creating vocabularies in sockeye.train.create_data_iters_and_vocabs
Issue -
State: open - Opened by iPRET 6 months ago
#1112 - Use `weights_only=True` when loading SockeyeModel parameters.
Pull Request -
State: closed - Opened by mjdenkowski 6 months ago
- 1 comment
#1111 - `accuracy` metric is no longer supported
Issue -
State: open - Opened by AmitMY 6 months ago
#1110 - Factors inference is slow (3 seconds/token) on A100 GPU
Issue -
State: open - Opened by AmitMY 6 months ago
- 6 comments
#1109 - Missing `--softmax-temperature T`
Issue -
State: open - Opened by AmitMY 6 months ago
- 1 comment
#1108 - Add alignment matrix learning
Pull Request -
State: closed - Opened by iPRET 6 months ago
- 4 comments
#1107 - Allow easier programmatic use by extracting code from `translate.py`
Pull Request -
State: closed - Opened by AmitMY 8 months ago
- 2 comments
#1106 - Out-of-memory when using target factors
Issue -
State: open - Opened by AmitMY 9 months ago
- 4 comments
#1105 - Guided alignments in Sockeye.
Issue -
State: open - Opened by tomsbergmanis 9 months ago
- 6 comments
#1104 - feat(evaluate): add signwriting evaluation metrics
Pull Request -
State: closed - Opened by AmitMY 10 months ago
- 2 comments
#1103 - Adding custom metrics
Issue -
State: open - Opened by AmitMY 10 months ago
- 3 comments
#1102 - IndexError when I try to continue training
Issue -
State: open - Opened by chrkell about 1 year ago
- 1 comment
#1101 - TracerWarning while translating but no error showing
Issue -
State: open - Opened by chrkell about 1 year ago
- 1 comment
#1100 - Update url for nightly builds.
Pull Request -
State: closed - Opened by fhieber about 1 year ago
#1099 - Update README.md w/ recent publications using Sockeye
Pull Request -
State: closed - Opened by fhieber about 1 year ago
#1098 - sockeye.utils.SockeyeError: Target sequences are not token-parallel:
Issue -
State: open - Opened by jingshu-liu about 1 year ago
- 1 comment
#1097 - Fine Tuning on Different Data size mismatch for embedding_target
Issue -
State: closed - Opened by AmitMY about 1 year ago
- 1 comment
#1096 - beam_search fix for running with torch.use_deterministic_algorithms(True)
Pull Request -
State: closed - Opened by Jehovan about 1 year ago
#1095 - `learning_rate_scheduler_type: inv-sqrt-decay` not reaching `initial_learning_rate`
Issue -
State: closed - Opened by SamuelLarkin about 1 year ago
- 3 comments
#1094 - Refactor inference optimizations
Pull Request -
State: closed - Opened by xingniu over 1 year ago
#1093 - Use fault tolerant symlink during training to handle temporary file system inconsistencies.
Pull Request -
State: closed - Opened by mjdenkowski over 1 year ago
#1092 - Automatic termination of training
Issue -
State: closed - Opened by SKT-T1-Thecai over 1 year ago
- 2 comments
#1091 - fix typos in testset names, allow symlinked datadir
Pull Request -
State: closed - Opened by thammegowda over 1 year ago
#1090 - pca + testset overlap scripts
Pull Request -
State: closed - Opened by thompsonb over 1 year ago
#1089 - How to finetune no-factor translation model with factors?
Issue -
State: closed - Opened by SKT-T1-Thecai over 1 year ago
- 1 comment
#1088 - ppl decreasing very slow
Issue -
State: closed - Opened by muramaso over 1 year ago
- 3 comments
#1087 - Fixes to blocking cross-attention
Pull Request -
State: closed - Opened by xingniu over 1 year ago
#1086 - Two fixes to SampleK
Pull Request -
State: closed - Opened by tdomhan over 1 year ago
#1085 - Add blocking cross-attention between decoder and encoded prepended tokens
Pull Request -
State: closed - Opened by xingniu almost 2 years ago
#1084 - RuntimeError: UNSUPPORTED DTYPE (sockeye-translate --use-cpu --dtype bfloat16)
Issue -
State: open - Opened by SamuelLarkin almost 2 years ago
#1083 - RuntimeError: "LayerNormKernelImpl" not implemented for 'Half' (sockeye==3.1.27)
Issue -
State: closed - Opened by SamuelLarkin almost 2 years ago
- 2 comments
#1082 - Modified factor decoding for automatic dubbing
Pull Request -
State: closed - Opened by Proyag almost 2 years ago
#1081 - UnicodeEncodeError: 'charmap' codec can't encode character
Issue -
State: closed - Opened by Verkolle almost 2 years ago
- 2 comments
#1080 - ValueError: setting an array element with a sequence.
Issue -
State: closed - Opened by Verkolle almost 2 years ago
- 5 comments
#1079 - BFloat16 Integration/System Test Update
Pull Request -
State: closed - Opened by mjdenkowski almost 2 years ago
#1078 - Specif or omit --shared-vocab consistently when training and preparing the data error
Issue -
State: closed - Opened by NLP-Peter almost 2 years ago
- 2 comments
#1077 - Enable bfloat16 inference and fix unrelated NumPy issue
Pull Request -
State: closed - Opened by mjdenkowski almost 2 years ago
#1076 - Code cleanup: refactoring, type checking, and formatting
Pull Request -
State: closed - Opened by mjdenkowski almost 2 years ago
#1075 - Update to sacrebleu 2.3.1
Pull Request -
State: closed - Opened by fhieber about 2 years ago
#1074 - Update to PyTorch 1.13 (+ 1.14 compatibility)
Pull Request -
State: closed - Opened by fhieber about 2 years ago
- 1 comment
#1073 - tok/sec throughput
Issue -
State: closed - Opened by vince62s about 2 years ago
- 3 comments
#1072 - OSError: libcudnn.so.7: cannot open shared object file: No such file or directory
Issue -
State: closed - Opened by nitin-bansal46 about 2 years ago
- 1 comment
#1071 - Update DeepSpeed checkpoint conversion to support newer DeepSpeed versions
Pull Request -
State: closed - Opened by mjdenkowski about 2 years ago
#1070 - How to avoid out of memory errors on large models
Issue -
State: closed - Opened by ejkitchen about 2 years ago
- 3 comments
#1069 - translation speed with quantization into int8
Issue -
State: closed - Opened by ZhenYangIACAS about 2 years ago
- 1 comment
#1068 - Change decoder softmax size logging level from info to debug.
Pull Request -
State: closed - Opened by mjdenkowski about 2 years ago
#1067 - AssertionError: If capturable=False, state_steps should not be CUDA tensors.
Issue -
State: closed - Opened by SamuelLarkin about 2 years ago
- 3 comments
#1066 - Add --tf32 device flag for transparent float32 acceleration
Pull Request -
State: closed - Opened by graehl about 2 years ago
- 5 comments
#1065 - BrokenPipeError: [Errno 32]
Issue -
State: closed - Opened by whher about 2 years ago
- 6 comments
#1064 - translator: send skip_nvs and nvs_thresh from args to ctor
Pull Request -
State: closed - Opened by graehl about 2 years ago
- 5 comments
#1063 - Train NVS using Sockeye
Issue -
State: closed - Opened by HossamAmer12 about 2 years ago
- 11 comments
#1062 - Add kNN-MT
Pull Request -
State: closed - Opened by shuoyangd about 2 years ago
- 12 comments
#1061 - Data streaming in Sockeye 3
Issue -
State: closed - Opened by tomsbergmanis about 2 years ago
- 2 comments
#1060 - Implement the Scones paper
Pull Request -
State: closed - Opened by jvamvas about 2 years ago
- 1 comment
#1059 - Add Support for DeepSpeed ZeRO Stage 1
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
- 2 comments
#1058 - Update README.md with some recent publications
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1057 - [Review #1051 First] Refactoring Ahead of Adding DeepSpeed Support
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
- 2 comments
#1056 - Update website with latest arxiv paper
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1055 - Out of memory exception after releasing GPUs in epoch 0
Issue -
State: closed - Opened by ejkitchen over 2 years ago
- 2 comments
#1054 - Guided alignments ala Garg et al.
Issue -
State: closed - Opened by tomsbergmanis over 2 years ago
- 7 comments
#1053 - About the Arxiv benchmark and the paper
Issue -
State: closed - Opened by vince62s over 2 years ago
- 13 comments
#1052 - Add Sockeye 3 paper to citations and BibTeX file
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1051 - Option to clamp layer outputs to min/max finite values for their dtypes
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1050 - Support offline model quantization
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1049 - How to plot attention matrix
Issue -
State: closed - Opened by etexaco123 over 2 years ago
- 1 comment
#1048 - Nbest reranking options using isometric translation criteria.
Pull Request -
State: closed - Opened by surafelml over 2 years ago
- 9 comments
#1047 - Fix data type of valid_length
Pull Request -
State: closed - Opened by hoangcuong2011 over 2 years ago
#1046 - Neural vocabulary selection.
Pull Request -
State: closed - Opened by tdomhan over 2 years ago
#1045 - Option to not reload the best training checkpoint when reducing the learning rate
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1044 - Fix scoring with batches of size 1 by avoid squeezing the batch dim
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1043 - Do not create incorrect checkpoints when resuming training with a fully trained model
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1042 - Ignore false-positive missing keys for traced modules when loading model parameters
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1041 - Update Readme.md with recent publications
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1040 - Model parameter keys are prefixed with "traced"
Issue -
State: closed - Opened by bricksdont over 2 years ago
- 10 comments
#1039 - Using SLURM, 2 nodes each with 4 GPUs [problem]
Issue -
State: closed - Opened by SamuelLarkin over 2 years ago
- 13 comments
#1038 - ValueError: Unsupported nproc_per_node value: 4(x2)
Issue -
State: closed - Opened by SamuelLarkin over 2 years ago
- 1 comment
#1037 - Add comment about "traced" naming convention
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1036 - Do not save parameters for traced modules.
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1035 - Optimized creation of repeated_state by using expand instead of repeat_interleave.
Pull Request -
State: closed - Opened by hoangcuong2011 over 2 years ago
- 3 comments
#1034 - I have challenges during sockeye training
Issue -
State: closed - Opened by etexaco123 over 2 years ago
- 1 comment
#1033 - Clarify use of Translator.batch_size in code
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1032 - Also trace SockeyeModel components when `inference_only == False` (includes CheckpointDecoder)
Pull Request -
State: closed - Opened by mjdenkowski over 2 years ago
#1031 - Moved offsetting of topk scores out of the (traced) TopK module.
Pull Request -
State: closed - Opened by fhieber over 2 years ago
Labels: enhancement
#1030 - I need some help with sockeye 3.1
Issue -
State: closed - Opened by etexaco123 over 2 years ago
- 1 comment
#1029 - Allow PyTorch 1.11 in requirements
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1028 - Code formatting & cleanup for beam_search.py and inference.py
Pull Request -
State: closed - Opened by fhieber over 2 years ago
#1027 - Vectors/Images/Video to Text Translation
Issue -
State: closed - Opened by AmitMY over 2 years ago
- 4 comments
#1026 - SockeyeError: Model(s) require 1 factors, but 2 given (through --input and --input-factors).
Issue -
State: closed - Opened by RamoramaInteractive over 2 years ago
- 1 comment
#1025 - Support target prefix with JSON format
Pull Request -
State: closed - Opened by hoangcuong2011 over 2 years ago
- 5 comments
Labels: feature, sockeye_3
#1024 - support JSON source prefix
Pull Request -
State: closed - Opened by hoangcuong2011 almost 3 years ago
- 4 comments
Labels: feature
#1023 - Faster prepare_source_length_mask using expand
Pull Request -
State: closed - Opened by fhieber almost 3 years ago
#1022 - Integration with huggingface
Issue -
State: closed - Opened by bricksdont almost 3 years ago
- 8 comments
#1021 - Allow torch==1.10.2, rename unit test file
Pull Request -
State: closed - Opened by fhieber almost 3 years ago
#1020 - Fixed GPU-based scoring by copying to cpu tensor first before converting to numpy.
Pull Request -
State: closed - Opened by fhieber almost 3 years ago
- 1 comment
Labels: bug
#1019 - Source-Factoring Translation not as expected
Issue -
State: closed - Opened by RamoramaInteractive almost 3 years ago
- 3 comments
#1018 - Added sacrebleu's TER metric to checkpoint decoder metrics.
Pull Request -
State: closed - Opened by fhieber almost 3 years ago
#1017 - Mismatch in arguments for training continuation. Differing arguments: validation_target, validation_source
Issue -
State: closed - Opened by RamoramaInteractive almost 3 years ago
- 2 comments