Release Notes (Archive)

Introducing Moderate (Beta)
Use Moderate (Beta) to classify harmful text across the following categories: profane, hate speech, violence, self-harm, sexual, sexual (non-consenual), harassment, spam, information hazard (e.g., pii). Moderate returns an array containing each category and its associated confidence score. Over the coming weeks, expect performance to improve significantly as we optimize the underlying model.

Model parameter now optional Our APIs no longer require a model to be specified. Each endpoint comes with great defaults. For more control, a model can still be specified by adding a model param in the request.

New Extremely Large
Our new and improved xlarge has better generation quality and a 4x faster prediction speed. This model now supports a maximum token length of 2048 tokens and frequency and presence penalties.

Updated Small, Medium, and Large Generation Models
Updated small, medium, and large models are more stable and resilient against abnormal inputs due to a FP16 quantization fix. We also fixed a bug in generation presence & frequency penalty, which will result in more effective penalties.

New & Improved Generation and Representation Models

We've retrained our small, medium, and large generation and representation models. Updated representation models now support contexts up to 4096 tokens (previously 1024 tokens). We recommend keeping text lengths below 512 tokens for optimal performance; for any text longer than 512 tokens, the text is spliced and the resulting embeddings of each component are then averaged and returned.

Classification Endpoint

Classification is now available via our classification endpoint. This endpoint is currently powered by our generation models (small and medium) and supports few-shot classification. We will be deprecating support for Choose Best by May 18th. To learn more about classification at Cohere check out the docs here.

New & Improved Generation Models

We’ve shipped updated small, medium, and large generation models. You’ll find significant improvements in performance that come from our newly assembled high quality dataset.

Finetuning is Generally Available

You no longer need to wait for Full Access approval to build your own custom finetuned generation or representation model. Upload your dataset and start seeing even better performance for your specific task.

Policy Updates

The Cohere team continues to be focused on improving our products and features to enable our customers to build powerful NLP solutions. To help reflect some of the changes in our product development and research process, we have updated our Terms of Use, Privacy Policy, and click-through SaaS Agreement. Please carefully read and review these updates. By continuing to use Cohere’s services, you acknowledge that you have read, understood, and consent to all of the changes. If you have any questions or concerns about these updates, please contact us at support@cohere.ai.

Extremely Large (Beta) Release

Our biggest and most performant generation model is now available. Extremely Large (Beta) outperforms our previous large model on a variety of downstream tasks including sentiment analysis, named entity recognition (NER) and common sense reasoning, as measured by our internal benchmarks. You can access Extremely Large (Beta) as xlarge-20220301. While in Beta, note that this model will have a maximum token length of 1024 tokens and maximum num_generations of 1.

Larger Representation Models

Representation Models are now available in the sizes of medium-20220217 and large-20220217 as well as an updated version of small-20220217. Our previous small model will be available as small-20211115. In addition, the maximum tokens length per text has increased from 512 to 1024. We recommend keeping text lengths below 128 tokens for optimal performance; for any text longer than 128 tokens, the text is spliced and the resulting embeddings of each component are then averaged and returned.

Representation Finetuning

Representation finetuning is now available in the dashboard. Upon uploading a .csv file in the format of [Examples,Labels] with a minimum of 200 examples, users will be able to finetune a baseline represenation model. For optimal finetunes, we recommend a minimum of 500 examples. Head on to Represenation Finetunes for more information.

Embeds Max Batch Size

The Embed endpoint's maximum number of strings per call has been lowered from 100 to 5.

Token Likelihood Endpoint

The Likelihood endpoint has been removed. To retrieve token likelihoods, use the return_likelihoods parameter of the Generate endpoint. Consequently, Likelihood is no longer a separate page in the Playground.

Generate Feedback

We’ve added a feedback button to the Generate playground. If you come across a generation that is below your expected quality, classify the result to help improve the Cohere Platform.

Versioning

The API version can be specified by setting the Cohere-Version header to the desired date, beginning with 2021-11-08.
The date must be a valid version.

Multiple Generations

Generate requests can use the num_generations parameter to specify how many generations should be returned.
This feature requires the Cohere-Version header to be set with a version of 2021-11-08 or higher.

Go SDK

The Cohere Go SDK is now live. Head on to API Reference for installation instructions.

Marine Life Model Update

We’ve adjusted our model offerings and bid farewell to the ocean animals. Instead, we’ll now provide access to three models: Small (previous seal), Medium (previous shark), and Large (previous orca). As a farewell to our ocean animals, we’ll donate $20,000 to an ocean wildlife foundation. Although code pointed toward our previous models won’t break, please update to our new offerings by December 1st.

Orca Model Update

Calls to orca will now be faster, produce better generations, and have the context size of 2048 tokens.

This update is automatic. Any code pointed towards the orca model will start seeing these improvements immediately.

New & Improved Generation and Embedding Models

otter (representation) is part of our new line of embedding models. Our tests show this line to run significantly faster and with improved performance on some tasks, and has an increased vector size of 1024 tokens. We suggest migrating all upcoming embedding usage to this model.

shark (generation) has significant advancements in speed and performance and now has double the capacity of tokens (2048). Any new calls to shark will be served by this new model.

Seal Model Update

The updated seal model runs faster with a noticeable increase in quality, shown by our internal tests. This model is generally more performant across all tasks, especially those involving common sense, reasoning, and reading comprehension.

This update is automatic. Any code pointed towards the seal model will start seeing these improvements immediately.

Choose Best Likelihoods

The Choose Best endpoint now returns scores, tokens, and token_log_likelihoods.

Finetuning Validation

Sample validation is now available in the dashboard. Upon uploading a .txt file, there is an option to "Review your Finetine" to validate a set of sample training data.

developer hub and documentation for Cohere Test!