Current Status

All services are operational. Occasional service interruptions may occur as we adjust system configuration.

Lab-hosted systems will be offline during the week of Oct 14th through 19th during upgrades to the cooling and power systems in Building 50.

Recent Updates

September 9th, 2024

• CBorg Nano model changed to Phi 3.5 with vision support - it can describe images and read text. VS Code ‘Continue’ support is now available in early beta for API users.

August 26th, 2024

• Lab-hosted models have been renamed with generic “CBorg” naming; underlying model has been replaced from Llama 3.1 to Mistral Large 2047 (licensed for non-commercial use). Added “Cborg Coder” (also based on Mistral Large with customized system message) and “Cborg Nano” (based on Microsoft Phi 3.5 - lightweight model for summarization and extraction tasks).

August 24th, 2024

• Google Gemini models are now available on the API service.

August 6th, 2024

• Low cost ChatGPT 4o-mini with regional deployment on Azure cloud. Implemented performance improvements to self-hosted chat models: Llama 3.1 and Command R+. New Lab-hosted embedding model lbl/nomic-embed-text now available.

July 31st, 2024

• The default temperature of models has been adjusted to 0.5. To customize the behavior of chat models please use a user preset.

July 30th, 2024

• CBORG Chat now supports RAG using CSV, TXT and PDF files.

July 25th, 2024

• Meta Llama 3 70B has been upgraded to Meta Llama 3.1 405B.

July 24th, 2024

• Access to Anthropic Claude through the API service is now working. ChatGPT-4o endpoint bandwidth was increased to accomodate user demand.

July 2nd, 2024

• Added a local API endpoint, https://api-local.cborg.lbl.gov which bypasses Cloudflare for local network clients.

June 24th, 2024

• Upgraded Claude 3.0 Sonnet to the latest Claude 3.5 Sonnet. Details: https://www.anthropic.com/news/claude-3-5-sonnet

June 17th, 2024

• Added LBNL-hosted embedding models e5-large-v2 and NV-Embed-v1, a large 4096-dimension embedding model with a leading position on the MTEB Leaderboard.

June 13th, 2024

• Added graceful failover to LBNL-hosted models in event of server offline
• Corrected incorrect configuration of cost-per-token setting in LBNL-hosted Command R+

June 12th, 2024

• LBNL-Hosted Llama-3 70B is now running on Nvidia H100 node for increased performance
• Custom chat icons have been provided for LBNL-hosted models

June 7th, 2024

• Removed support for ChatGPT 4 (legacy model), replaced by GPT-4o
• Removed support for Google Gemini 1.0 Pro, replaced by Gemini 1.5 Flash and Gemini 1.5 Pro
• Adjusted context window for LBNL-hosted Command R+ to 80K tokens per system memory constraints
• Adjusted maximum context length for commercial models to reasonable limits for chat use to control costs
• Improved model selection drop-down list with model descriptions

June 6th, 2024

• Added support for ChatGPT 4o
• Added support for Google Gemini 1.5 Flash and Google Gemini 1.5 Pro
• Fixed Anthropic Claude endpoints
• Added support for Anthropic Claude Opus