🐦 junelark
Contact sales

Buy state-of-the-art language model weights.

  • Models range from 1B to 1T params, trained on 1T to 10T tokens
  • Deal sizes from $500k to $30M, flexible timing
  • Startups can pay with equity
  • Enterprise grade support
Contact sales

MODEL CAPABILITIES

Capabilities menu:

  • Models from 1B to 1T parameters
  • Training datasets from 1T to 10T tokens
  • Multimodal input, multimodal output
  • Continuous training on new web data
  • Real-time RLHF from user feedback
  • Context windows from 8k to 10M tokens
  • Code pretraining
  • Retrieval from large corpuses
  • Low latency ("turbo")
  • We can develop novel modalities as needed, music, voice, financial data

* if there are capabilities you need and don't see listed, contact us

Built by ex-OpenAI, Harvard, Google Brain, FAIR, Deepmind, Cohere, MIT, Stanford

Be state-of-the-art.

  • State-of-the-art applications require access to the weights
  • Creative applications need creative fine-tuning
  • Develop new capabilities with frontier research on top of a base model
  • Push the limits with neural hackery
  • Use an LM as a base policy for RL

On-prem or on-device

  • On-prem is more secure, private, and reliable
  • Lower-latency than calling an api
  • Full control of the production environment

Integration and deployment

  • We provide inference infrastructure
  • Fine-tuning support on proprietary datasets
  • RLHF support for tuning chatbot responses

Novel modalities

  • We can patch in custom modalities as needed:
  • Music
  • Voice
  • Financial data
  • Robotics control signals

RESEARCH SERVICE

AI Research-as-a-Service

  • We can develop novel AI capabilities for specific use-cases
  • Language models are successful because they have scalable datasets and scalable architectures
  • Other tasks can be massively scaled up to trillions of parameters
  • We are excited to help with scale-up projects in other domains as the techniques underlying language models make their way to other industries
Contact sales

Tasks that can be scaled up

  • Reinforcement learning
  • Self-driving
  • CNC simulation
  • Financial modeling
  • Retail demand forecasting
  • Chip design
  • CAD design
🐦 junelark

2023