flairNLP / fabricator

[EMNLP 2023 Demo] fabricator - annotating and generating datasets with large language models.

flairNLP/fabricator Issues

Failing CI pytest
Closed 8 months ago
Supporting other open-source LLMs
Closed 8 months ago2
Remove spacy dependency
Closed 10 months ago
Initial version for pypi package
Closed 10 months ago
Naming of sampling strategies
Updated 10 months ago1
after accepted: create pypi package
Closed 10 months ago1
Cannot import DatasetGenerator
Closed 10 months ago1
Large Git Pack Files
Closed 10 months ago2
Implement ProGen approach
Updated a year ago2
Add Fabriator Predictions to Wrench Datasets
Updated a year ago
Guided Generation for Syntax dependent tasks
Updated a year ago
Idea on how to structure generation / annotation
Updated a year ago1
Add predicted labels to CoNLL-03 dataset
Updated a year ago1
Improve fewshot sampling naming convention
Closed a year ago
Rename to Fabricator
Closed a year ago
improve logo + fix width in readme
Updated a year ago
Split haystack dependencies
Closed a year ago2
for evaluation: compare against baseline that just k times repeats the few shot examples
Updated a year ago
severe: class_encode_column does not convert to same original int labels
Updated a year ago
Reduce dependencies
Closed a year ago1
enable user to provide 1-sentence description for each label, which will be shown at the beginning of the prompt to the LLM
Closed a year ago2
Refactorings for submission
Closed a year ago
Custom Prefixes for data column and few shot column for prompt
Closed a year ago2
Rename source directory to "src"
Closed a year ago
More log infos
Closed a year ago
cost reduction: generate and annotate within a single prompt (instead of first generating and then annotating)
Updated a year ago
execution speed degrades with increasing sizes of datasets to be annotated
Closed a year ago3
Ensure robust and fault tolerant generation
Closed a year ago
Make it possible to use all prompts with text inputs rather than label IDs
Closed a year ago
How to properly differentiate between generate unlabeled and annotate unlabeled data?
Closed a year ago2
Naming convention for DataPoints / Prompts
Closed a year ago
Sampling method that covers all labels
Closed a year ago