v0.1.2

What's Changed

• 增加复现实验的中文文档 by @aJupyter in #83
• fix bug of tiktoken.encoding_for_mode by @toughhou in #91
• [Edit] add framework specification in trace by @yeahjack in #92
• update retriever by @plageon in #93

New Contributors

• @aJupyter made their first contribution in #83
• @toughhou made their first contribution in #91
• @yeahjack made their first contribution in #92
• @plageon made their first contribution in #93

Full Changelog: v0.1.1...v0.1.2

v0.1.1: Trace, Spring, IRCoT Integration

New Features

• Added support for the Azure OpenAI model. (cbd4c8d)
• Added a new method: trace. This feature was implemented twice, suggesting it might have been enhanced or corrected in the second commit. (b9d6fa1, 7a728a2)
• Integrated Sentence Transformers into the retriever module. (12e330e)

New Models

• Optimized generator settings specifically for the Baichuan model. (7e0f124)

Bug Fixes

• Fixed a bug in the retrieval cache saving mechanism. (18a727e, 45662e3)
• Fixed an error in the evaluation metrics. (d0f403e)
• Resolved issues with the cache manager. (8277011)
• Fixed a bug in the sentence transformer encoder. (b0fdfbb)
• Resolved a bug in single GPU inference. (982d433)
• Fixed a bug in BM25 functionality. (5009089)
• Resolved a bug in the loading of T5-based models in DataParallel. (6666b4a)
• Fixed a bug related to the max token setting in the EncoderDecoder generator. (96b20c8)

Enhancements

• Optimized the implementation of the generator. (d7a7136)
• Optimized the detection of generator types. (ed7ef8e, 72c831d)
• Optimized the refiner's memory usage. (1984ca8)
• Optimized generator settings for better performance. (7e0f124)
• Modified the LoRA loading of the generator under VLLM setting. (c55ea39)
• Increased support for multi-choice datasets in metrics. (95d9fcc)
• Modified data path settings in the evaluator. (70e739e)
• Changed the metric name from sub_em to acc. (15de563)
• Changed sample data to general knowledge. (3d20a75)
• Updated default configuration settings. (1acce56)

Documentation & Code Formatting

• Added guidance for setting up configurations. (37fc012)
• Updated the README file multiple times for various improvements, including adding installation instructions, fixing typos, and providing more detailed documentation.
• Formatted code files for better readability. (785a901)
• Added an introduction for beginners in Korean. (c25f16e)

New Contributors

• @PeterGriffinJin made their first contribution in #23
• @yifeikong made their first contribution in #24
• @kerrychu made their first contribution in #39
• @VSteinborn made their first contribution in #52
• @oneonlee made their first contribution in #57

Full Changelog: v0.1.0...v0.1.1

V0.1.0: Release FlashRAG

Features

• Implement 12 RAG algorithms
• Various RAG-Components
• Release 32 FlashRAG datasets and wiki corpus
• Support vllm, FastChat, faiss

New Contributors

• @BUAADreamer made their first contribution in #6

Full Changelog: https://github.com/RUC-NLPIR/FlashRAG/commits/v0.1.0