microsoft/KBLaM
A method for augmenting large language models with knowledge bases without external retrieval modules, presented at ICLR 2025.
Not currently ranked — collecting fresh signals.
star history
KBLaM provides an alternative to retrieval-augmented generation that eliminates external retrieval modules entirely. Instead, it integrates external knowledge directly into the LLM architecture, achieving linear computational scaling with knowledge base size rather than the quadratic scaling of in-context learning approaches. The implementation supports popular Hugging Face models including Llama-3 and Phi-3, and includes tools for generating synthetic knowledge bases and embedding them for use with the augmented language models.
Frequently asked
- What is microsoft/KBLaM?
- A method for augmenting large language models with knowledge bases without external retrieval modules, presented at ICLR 2025.
- Is KBLaM open source?
- Yes — microsoft/KBLaM is open source, released under the MIT license.
- What language is KBLaM written in?
- microsoft/KBLaM is primarily written in Jupyter Notebook.
- How popular is KBLaM?
- microsoft/KBLaM has 1.4k stars on GitHub.
- Where can I find KBLaM?
- microsoft/KBLaM is on GitHub at https://github.com/microsoft/KBLaM.