(docs)Update meta description

[lampa-da]

This PR introduces a range of updates to the Cohere platform, including new models, improved performance, and enhanced capabilities.

• New Models:
  • medium-20220217, large-20220217, and an updated small-20220217 representation models are introduced, offering improved performance and flexibility.
  • The xlarge-20220301 model is released as a beta version, providing better performance on various NLP tasks.
  • A new classification endpoint is added, powered by generation models, enabling flexible few-shot classification.
  • The multilingual-22-12 model is introduced for multilingual text understanding and language detection, enabling easy cross-language searches.
  • The Command R+ model is released, optimized for conversational interaction and long-context tasks.
  • The Rerank 3 model is launched, offering improved performance and speed for ranking tasks with a context length of up to 4096 tokens.
• Improved Performance:
  • Updated generation models (small, medium, and large) provide enhanced performance due to a high-quality dataset.
  • The xlarge model is improved, offering faster and better generations with a maximum token length of 2048 tokens.
  • Updated small, medium, and large models are more stable and resilient against abnormal inputs due to a FP16 quantization fix.
  • Retrained small, medium, and large generation and representation models support longer contexts of up to 4096 tokens.
  • The Moderate (Beta) model is introduced for harmful text classification, with improved performance expected over time.
  • The Co.classify endpoint now supports multilingual text classification with the multilingual-22-12 model.
• Enhanced Capabilities:
  • Fine-tuning is now generally available, allowing users to fine-tune models on custom datasets without Full Access approval.
  • APIs offer greater flexibility with optional model specifications and default endpoint settings.
  • The Co.classify endpoint now serves few-shot classification tasks using embeddings from the Representational model.
  • The logit_bias parameter is added to generative models to prevent unwanted tokens and incentivize desired tokens.
  • The Command model is updated with improved code, conversational capabilities, and reasoning skills.
  • The co.chat() beta is released with RAG for grounded AI output, updated command models, and new rate limits.
  • The Python SDK is updated with fine-tuning capabilities, new local tokenizer defaults, and required 'model' fields.
• Other Updates:
  • The maximum tokens length per text is increased from 512 to 1024 for representation models.
  • The xlarge model now supports frequency and presence penalties.
  • A bug in the generation presence and frequency penalty is fixed for more effective results.
  • The developer documentation is updated with interactive tutorials and a refreshed UI.
  • Model names are updated for simplicity and consistency, and all documentation is updated accordingly.
  • The maximum number of input documents for co.rerank is updated, with an error endpoint included.
  • New free trial keys, custom model features, reduced pricing, and improved team sharing options are introduced.

[github-actions]

🌿 Preview your docs: https://cohere-preview-ba230e91-2bc3-44c3-ad4d-9f93c963958f.docs.buildwithfern.com

[github-actions]

🌿 Preview your docs: https://cohere-preview-3825068c-5edc-4162-9fd4-cff438db7de6.docs.buildwithfern.com