Hello, and welcome to this week’s changelog update! Let’s dive into what’s new:
-
Tenstorrent Private Preview: New One-Click Models & Fix for Sporadic Card Init/Reset Delays
We’ve added one-click models ready to deploy for Tenstorrent Instances, including Llama 3.1-8B Instruct, Llama 3.2-1B and 3B Instruct, and Qwen 2.5-7B Instruct. All of these models are served using vLLM (Tenstorrent).
We’ve also fixed an issue that sometimes caused sporadic card init and reset delays.
Looking to access on-demand tenstorrent hardware with Koyeb? Request access
-
Improved Errors on Deployment Failures
We’ve made deployment failure messages more informative to help you troubleshoot issues faster. You’ll now see clearer error messages when:
- The executable file used to launch the application is not found
- The container image download fails because the image no longer exists
Previously, these issues resulted in a generic error, making them difficult to diagnose and resolve.
-
Control Panel: Improvements & Fixes
We’ve released a bunch of improvements and fixes in the control panel, here is the details:
- Display a clearer message when deployments are in a pending state due to the plan’s concurrent limit being reached
- Prevent creating a new deployment when a referenced value used in interpolation does not exist for environment variables and config files
- Fixed a display bug where the target CPU usage autoscaling criteria was selected despite the min and max instances being equal.
- Prevent enabling scale-to-zero for services using volumes
-
Meet the team at AI Rabbit Hole in the Bay Area next week!
Next week, we’ll be in the Bay Area for several exciting events:
- April 23: Join us for Welcome Drinks the night before the AI Rabbit Hole conference! We’d love to see you there!
- April 24 AI Rabbit Hole in SF: Catch our panel discussion on deploying AI agents, swing by our booth to chat with the team, pick up exclusive swag, and join us for lunch!
