Language model tuning in conversational artificial intelligence systems and applications
AssigneeNVIDIA Corporation
InventorsYi Dong, Xianchao Wu, Yi Fen Zenodia Charpy
Life of this patent
Estimating the term…
Abstract
Disclosed are systems and techniques that may generate prompts for language models. The techniques include obtaining a first dataset and a second dataset and training a hierarchical virtual token generator (VTG) model to generate a large language model (LLM) input prompt. Training the hierarchical VTG includes training, based on the first dataset, a first VTG to output a first virtual token and training, based on the second dataset, a second VTG to output a second virtual token embedding. The generated LLM input prompt includes the first virtual token embedding and the second virtual token embedding.
Claims20
A method comprising:
- training a hierarchical virtual token generator (VTG) model to generate a large language model (LLM) input prompt, the training comprising:
- training, based at least on a first dataset, a first VTG to output a first virtual token embedding;
- and
- training, based at least on a second dataset different from the first dataset, a second VTG to output a second virtual token embedding;
- wherein, during the training of the hierarchical VTG model, the LLM input prompt comprises the first virtual token embedding and the second virtual token embedding.
A method comprising:
- generating a large language model (LLM) input prompt using a hierarchical virtual token generator (VTG) model, wherein the hierarchical VTG model comprises:
- a first VTG to generate a first virtual token embedding, and
- a second VTG to generate a second virtual token embedding;
- and
- performing a natural language processing (NLP) task based at least on providing the LLM input prompt as input to an LLM, the LLM input prompt including the first virtual token embedding and the second virtual token embedding.
A system comprising:
- one or more processors to:
- generate a large language model (LLM) input prompt using a hierarchical virtual token generator (VTG) model, the hierarchical VTG model comprising:
- a first VTG to generate a first virtual token embedding, and
- a second VTG to generate a second virtual token embedding;
- and
- perform a natural language processing (NLP) task based at least on an LLM processing the LLM input prompt including the first virtual token embedding and the second virtual token embedding.
Dependencies are inferred from the claim text.