v1.5.7 Update Live: GPU Fixes, Faster Responses, and More

Patch v1.5.7 has been released

GPU Improvements

Fixed issues affecting GPU mode on some AMD and certain Nvidia setups.

GPU mode should now function as expected, and GPU acceleration for vision has been added as an optional setting for improved performance.

Vulkan support has also been implemented, improving compatibility across a wider range of systems and hardware configurations.

When enabling "Use GPU" Mode and/or GPU Vision (Beta), CielChan will restart by design to apply the changes. Once relaunched, these settings will carry over into that next session. You can choose to have GPU use for either text generations, vision, or both.

Performance Improvements

Further improved overall generation speed and responsiveness. If you were experiencing slower replies, this update should help improve consistency across more systems.

For CPU-only usage with a 3B model, response times can vary depending on hardware, but a few seconds per response is generally expected. If responses are taking significantly longer, there may be system-specific bottlenecks affecting performance.

When sending your first vision prompt in a session, it may take noticeably longer than subsequent requests, as the vision model requires a brief warm-up before responding at normal speeds.

Similarly, when launching CielChan after the app has been closed, it may take a moment to prepare the environment and load your assistant. Generation speeds will increase after the initial response as everything becomes fully initialized.

UI and Onboarding Improvements

A new command button has been added to the vertical bar, providing a set of simple example commands for newer users to try.

An updated onboarding description now appears when starting the app from scratch, giving additional context on how CielChan works and what to expect.

General UI improvements and refinements have also been made to provide a smoother overall experience.

Assistant Improvements

Added a preliminary ability for CielChan to read and reference her own help and settings, allowing for more guided responses and improved self-awareness when assisting users.

Localization Improvements

Further refinements and fixes have been made to localization, improving consistency and reducing edge case issues across supported languages.

Fixed remaining localization inconsistencies, including incorrect or missing words/keys for Russian, Dutch, and other languages, along with improvements to language toggle behavior.

Minigame Improvements

Improved CielChan’s banter and interaction during minigames, including Fleet Duel, Tic Tac Toe, and Territory Control, for a more engaging experience.

(You can enable TTS while playing minigames in the top-left corner of the minigame panel.)

Fixes and Refinements

Additional fixes and improvements have been made across various systems to improve overall stability and reliability. Custom Model Support Reminder

You can load your own LLM GGUF models for experimentation or if your system supports it.

To do this, place your GGUF model file into the CielChan local steam directory: CielChan → CielChan Desktop AI_Data → StreamingAssets

Once added, launch CielChan and your model will appear in the dropdown for selection.

If you run into any issues with the update, please post them in the discussion boards so they can be investigated.

Thanks everyone.