Guides
Log In
Guides

Model Library

NuPIC offers a wide range of pre-trained and optimized models for different use cases. We include both non-generative and generative models. If you need customization for your specific use case, you can fine-tune a model with the NuPIC Training Module. You can also choose to import your own model from outside NuPIC if it has a supported architecture.

📘

Which model?

We recommend choosing an appropriate model based on your use case, accuracy and performance requirements, and available resources. Find tips on how to select the right model here.

Non-Generative Models

Model NameDescription
bert.baseAn unoptimized model from Hugging Face for benchmarking.
nupic-bert.ab.large-v0An optimized model for embedding texts while offering exceptional adaptability to different NLP tasks.
nupic-bert.ab.base-v0A smaller version of nupic-bert.ab.large-v0, ideal for real-time applications.
nupic-sbert.large-v1An optimized model that excels in sentence-level tasks.
nupic-sbert.base-v3A smaller, much faster version of nupic-sbert.large-v1. Ideal for real-time applications.
nupic-sbert.ab.large-v2An optimized model that excels in sentence-level tasks with higher pre-training and finetuning accuracy.
nupic-sbert.ab.base-v5A smaller, much faster version of nupic-sbert.ab.large-v2. Ideal for real-time applications

Generative Models

Model NameDescription
nupic-gpt.7b-corti.v0NuPIC's generative AI model, specifically optimized using neuroscience principles.
nupic-gpt.7b-dendi.v0NuPIC's generative AI model, specifically optimized using neuroscience principles.
gemma.it.2bA smaller, faster generative AI model. Needs to be downloaded separately.
gemma2.it.9bArchitectural and training enhancements over Gemma, in a new size class. Needs to be downloaded separately .
llama2.chat.7bA generative AI model designed for broad NLP applications. Needs to be downloaded separately.
llama3.chat.8bA generative AI model with performance and accuracy enhancements over Llama 2. Needs to be downloaded separately .
zephyr.beta.7bA generative AI model with high efficiency and adaptability in specific domain tasks.

Model Sizing

The size of a model can affect its performance, speed, and resource requirements.

  • Base (e.g. nupic-bert.ab.base-v0): These BERT-base equivalent models are relatively lightweight, and are therefore faster to run and require less compute. They are ideal for real-time applications, tasks that require low-latency inference, or deployment on edge.
  • Large (e.g. nupic-bert.ab.large-v0):These models are roughly twice as large as their base counterparts. These models, with their increased capacity, can understand more complex structures and nuances in data, which typically comes with an increase in accuracy.
  • GPT sizes (e.g. nupic-gpt.7b-corti.v0): GPT models have their sizes expressed in billions of parameters in their names. Performance, accuracy and compute characteristics of large language also vary with size, as explained above.

Tokenizers

  • With Tokenizer: Each model comes with an integrated tokenizer, making it able to accept raw text inputs.
  • Without Tokenizer (-wotokenizer): You can use the models without a tokenizer. You will need to tokenize the text data separately in the correct format before using them.

Output Streaming

Output streaming allows GPT models to print their generated outputs one token at a time. This helps to reduce perceived latency, which improves the overall user experience. Output streaming is particularly useful in use cases involving real-time user interactions, such as chatbots. NuPIC models that support output streaming are denoted by the .streaming suffix.

Can't find exactly what you need?

You can use one of the models from the Model Library as a starting point to fine-tune a model with your own data! You can also do the same by importing your own model from outside NuPIC.