Cohere trains massive language models and puts them behind a simple API. We handle the complexities of collecting massive amounts of text data, evolving neural network architectures, distributed training, and serving models around the clock.
Cohere offers three baseline models, each with different amounts of power (i.e., parameters): small, medium, and large. Our Embed and Generate endpoints can be used to implement these models across a variety of different tasks.
Our models gain a rich understanding of language and the world by reading text across billions of web pages.
Finetune our models to improve performance on downstream tasks or teach the model large quantities of information that cannot be extracted with few-shot learning or prompt engineering.
Access language generation and model representations under one API.
We handle of the complexities of collecting massive amounts of text data, architectures, distributed training, and serving models around the clock.
Return textual embeddings that capture semantic information about text.
import cohereco = cohere.CohereClient('YOUR_API_KEY')emb = co.embed(model='baseline-small',texts=['This album is not that great.', 'Their old stuff was better.'])if emotion_classifier.predict(emb) < 0.5:print('They did not like it')
Produce realistic text conditioned on a given input across a variety of tasks.
import cohereco = cohere.CohereClient('YOUR_API_KEY')predictions = co.generate(model='baseline-medium',prompt='Rephrase the following sentence:\n''1. When does the show start?\n''2. What time is the concert?\n''3. When is the beginning of the set?\n',num_tokens=50)