v3.1.0

What's Changed

• Update llava.py by @mosheber in #54
• Remove indexing function by @mosheber in #55
• IPEX benchmarking fix by @peteriz in #58
• Removing Handlers with Phi3.5 Suppport by @mosheber in #59
• replaced list[str] with List[str] by @mosheber in #67
• Adding files for multi modal pipeline by @mosheber in #68
• Lazy initialization of OVModel by @danielfleischer in #66
• update protobuf version to 5.28.3 by @mosheber in #70
• Update one link in nutrition_data.json by @bilgeyucel in #72

New Contributors

• @bilgeyucel made their first contribution in #72

Full Changelog: v3.0.2...v3.1.0

v3.0.2

What's Changed

• Fix IPEX embedders performance by @peteriz in #52
• Fix support python versions by @peteriz in #53

Full Changelog: v3.0.1...v3.0.2

v3.0.1

What's Changed

• Gaudi Generator by @mosheber in #50
• Adding pypi packaging support by @peteriz in #51

Full Changelog: v3.0...v3.0.1

v3.0.0

Compatibility with Haystack v2

• ⚡ All our classes are now compatible with 🤖 Haystack v2, including the example notebooks and yaml pipeline configurations.
• 💻 We based our demos on the Chainlit UI library; examples include RAG chat with multi-modality! 🖼️

❤️ Feel free to report any issue, bug or question!

v2.0.0

fastRAG 2.0: Let's do RAG Efficiently 🔥

fastRAG 2.0 includes new highly-anticipated efficiency-oriented components, an updated chat-like demo experience with multi-modality and improvements to existing components.

The library now utilizes efficient Intel optimizations using Intel extensions for PyTorch (IPEX), 🤗 Optimum Intel and 🤗 Optimum-Habana for running as optimal as possible on Intel® Xeon® Processors and Intel® Gaudi® AI accelerators.

🚀 Intel Habana Gaudi 1 and Gaudi 2 Support

fastRAG is the first RAG framework to support Habana Gaudi accelerators for running LLMs efficiently; more details here.

🌀 Running LLMs with the ONNX Runtime and LlamaCPP Backends

Added support to run quantized LLMs on ONNX runtime and LlamaCPP for higher efficiency and speed for all your RAG pipelines.

⚡ CPU Efficient Embedders

We added support running bi-encoder embedders and cross-encoder ranker as efficiently as possible on Intel CPUs using Intel optimized software.

We integrated the optimized embedders into the following two components:

• QuantizedBiEncoderRanker - bi-encoder rankers; encodes the documents provided in the input and re-orders according to query similarity.
• QuantizedBiEncoderRetriever - bi-encoder retriever; encodes documents into vectors given a vectors store engine.

⏳ REPLUG

An implementation of REPLUG, an advanced technique for ensemble prompting of retrieved documents, processing them in parallel and combining their next token predictions for better results.

🏆 New Demos

We updated our demos (and demo page) to include two new demos that depict a chat-like experience plus fusing multi-modality RAG.

🐠 Enhancements

• Added documentation for most models and components, containing examples and notebooks ready to run!
• Support for the Fusion-in-Decoder (FiD) model using a dedicated invocation layer.
• Various bug fixes and compatibility updates supporting the Haystack framework.

Full Changelog: v1.3.0...v2.0

v1.3.0

What's Changed

• ColBERT Upstream Updates by @danielfleischer in #19

Full Changelog: v1.2.1...v1.3.0

v1.2.1

What's Changed

• Update plaid_colbert_pipeline.ipynb by @mosheber in #17
• Update colbert.py by @mosheber in #18

Full Changelog: v1.2.0...v1.2.1

v1.2.0: New: Retrieval Augmented Generation with LLM

Retrieval Augmented Generation with LLM Demo (#16)

- Added a new RAG + prompt + LLM UI (demo).
- Added an example config and notebook.
- Updated main README with "updates" sub-section.
- Updated `run_demo.py` to include all the options to run a demo (UI, UI + service, UI + <user_defined_service>)