Arcee Overview
At Arcee, we believe in a world where organizations are building and maintaining their own AI models on top of open-source general intelligence. Unlike other AI companies, we follow a few core tenets:
Extending general intelligence through the parameter space of the model is the most effective way to adapt intelligence - (as opposed to techniques like prompting + RAG)
This entails that our customers' ownership of model parameters is a cornerstone of our system, which should be built on the open-source general checkpoint of the day.
We happen to believe one of the best ways to extend the open source general pretrained checkpoint is through model merging, where the parameters of two language models are combined. We offer a hosted and deployable platform to train and merge language models that are built on an open merging core, mergekit
Our platform adaptation system involves routines and GPUs to perform:
Continuous Pretraining - injects new knowledge to your language model by running next token prediction text on proprietary data.
Alignment - steers your model's response style
Merging - combines models of different capacities into one model
We also believe that no AI platform solves the nuance of any problem - thus we encourage you to train and merge models within and without the Arcee platform.
You can access the hosted version of these tools at app.arcee.ai