Two Integration Approaches for Bitext NAMER and LLMs
There are two major approaches to integrating Bitext’s NER system with an LLM like GPT or Llama:
- Pre-processing the enter textual content:
On this method, entities are annotated utilizing the NER system earlier than feeding the pre-annotated textual content to the LLM as a part of the enter immediate or context. That is notably helpful for bigger programs, the place explicitly connecting entities to present information graphs or databases is advantageous. This methodology is appropriate with just about any language mannequin.
- Mannequin-driven integration:
Right here, the LLM is configured to name the NER system immediately when wanted. This method requires an LLM platform that helps exterior API calls or “perform calling,” akin to GPT, Jamba, Mistral, and Llama. This methodology is good to be used circumstances the place finish customers work together immediately with the LLM, and the mannequin orchestrates a workflow based mostly on person enter.
Leveraging Knowledge from Bitext NAMER
The info generated by Bitext’s NAMER system could be utilized in numerous methods, together with:
- Producing an entity checklist with metadata to be used as a content material library.
- Direct integration into an index or information base.
- Sustaining Bitext NAMER output as a separate file for on-demand entry by analysts, researchers, or investigators.
