April 2023 Summaries

Databricks Dolly v2 12b is a 12 billion parameter instruction-tuned casual language model derived from EleutherAI's pythia-12b and pretrained on The Pile and GPT-J's pretraining corpus. This model is optimized using the open-source databricks-dolly-15k instruction-following dataset and can be deployed through the DeepInfra web dashboard or API, with the deployment triggered upon the first inference request. Inference requests can be made via a REST API, and usage is priced at $0.0005 per second, running on Nvidia A100 cards. DeepInfra provides a fully managed GPU infrastructure for scalable model hosting, and users can access further documentation and support through their website and Discord server.

Databricks Dolly v2 12b is a 12 billion parameter instruction-tuned casual language model based on EleutherAI's pythia-12b, pretrained on The Pile and GPT-J's corpus, and fine-tuned using the open-source databricks-dolly-15k dataset. DeepInfra facilitates the deployment and inference of this model through its platform, requiring users to obtain an API key for access. The model can be deployed via the web dashboard or API, with inference requests charged per execution time, running on Nvidia A100 cards. Additionally, DeepInfra integrates with LlamaIndex for document indexing and searching, and offers various other AI models and tools, available via their platform.

DeepInfra offers an API service that enables users to utilize OpenAI's Whisper, a Speech-To-Text model, for converting audio files into text with timestamp segmentation by sentence or word. Users need to create an API key through their DeepInfra account to authenticate requests, and the service supports various model sizes and language variants. The platform provides a REST API endpoint for using Whisper's speech recognition capabilities, with detailed documentation available for additional parameters and usage examples. DeepInfra emphasizes its affordable AI hosting with a fully managed GPU infrastructure, promising enterprise-grade uptime, and invites users to contact their sales team for further inquiries or reach out via their Discord server for support.