Cooperation on building semantics-aware large language models (LLMs)

Short summary A German SME is looking for a partner experienced in training, verifying and testing LLMs with natural language data. They deliver training and test data with high-quality natural language processing (NLP) annotation. The SME expects a partner with resources and skills to train LLMs and eventually to provisioning them to end customers. The partnership should be technological as well as commercial.

Full description The German SME has a deep knowledge and experience in computational linguistics and its employees have many patents and publications on software development and architecture in this field. Based on these assets, the company has developed a unique method to provide annotated training and test data for LLMs and other NLP applications, where these high-quality annotations enable an advanced compositional semantics for LLMs. The German company is looking for a partner experienced in training, verifying and testing LLMs specifically so-called transformer models with natural language data. Such models, e.g. Google (Bard) or Chat GPT (OpenAI) have become very popular recently. For the partner it would be beneficial but not necessary to work with end customers on NLP projects, be it in a production or services scenario. The partner would be enabled to provide an advanced NLP technology within their products or services. This is because there are studies showing that LLM effectiveness, as well as controllability and domain knowledge, can be substantially enhanced through the annotated data provisioned by the German SME. Partnership would include joint training of LLMs on a marketable domain, e.g. from a customer project, such that the LLM and, eventually, an NLP application based on it can be jointly offered to customers. Since such offering would solve a key business communication problem for many customer companies/organizations, this partnership can lead to significant success. Furthermore, the SME would be interested in joining a consortium for a Horizon Europe project under Cluster 4: Digital, Industry and Space, with the perspective of joint provisioning and marketing of such advanced LLMs to customers.

Technical Specification or Expertise Sought The partner should have access to an advanced server network and expertise in managing this network for a highly ambitious IT operation, which includes distributed computation of a LLM on server nodes with decent RAM requirements.

Contact / source: NEXT EEN Widgets (europa.eu)