The IMPACT-IBM team has developed a system called INDUS that is capable of achieving superior performance on various benchmarks related to biomedical tasks, scientific question-answering, and Earth science entity recognition. This was achieved by providing INDUS with domain-specific vocabulary, which allowed it to outperform open, non-domain specific LLMs in these areas.
One of the key strengths of INDUS is its ability to handle diverse linguistic tasks and retrieval augmented generation. This means that the system can process questions from researchers, retrieve relevant documents, and generate answers to these questions more effectively than other models.
To cater to applications that require low latency, the team also developed smaller and faster versions of both the encoder and sentence transformer models used in INDUS. This ensures that the system can perform efficiently in scenarios where response times are critical.
