0.3.6

This build updates the SDK to support updated routes.

When installing this update, please ensure you have also updated the API to at least 0.2.22

Full Changelog: v0.3.5...v0.3.6

0.3.5

Patch release to fix an issue that was breaking the Windows installer.

0.3.4

What's Changed

• Highly requested feature: stop current generation. You can now stop a generation that is in progress.
• Templated Prompts Page now lets you create your own prompt templates
• Improvements & bug fixes to Templated Prompts Page
• API page is simplified a tiny bit
• update deps + fix some TS by @rohannair in #76
• Revert "update deps + fix some TS" by @dadmobile in #78

New Contributors

• @rohannair made their first contribution in #76

Full Changelog: v0.3.3...v0.3.4

0.3.3

Big Changes:

• Addition of "Templated Prompts" page in the Interact section of the app
• Completions and Templated Prompt now stream text
• All requests to the LLM are logged and shown in the new Prompt Logs page
• Allow manual setting of stop strings in the Chat interface

More details:

• Improved error reporting during installation
• Improved error reporting when a model download fails
• Do not allow "eject" while a model is running
• Improvements to Interact (chat) page UI and code
• Hide some inference settting knobs in the Interact page and put them behind a "All Generation Settings" section
• Show context window in model details page
• Check and update Python dependencies when upgrading the API

Full changelog: v0.3.2...v0.3.3

0.3.2

Patch to fix unresponsive Step 1 on Windows Local Connection install.

0.3.1

• Improve RAG error display

0.3.0

Initial preview of Retrieval Augmented Generation and Windows support.

0.2.14

• Improved, minimalistic UI
• Regenerate button: ability to regenerate last message from LLM
• Save the last experiment you were on, per server
• Fix to local auto-installer

Full Changelog: v0.2.13...v0.2.14

0.2.13

• Uses the new expectation that conda is installed at ~/.transformerlab/miniconda3

0.2.6

• Updated local installer -- now only works with miniconda installed at ~/miniconda3
• Fixes numerous issues looking for conda
• Default inference temperature is set to 0.7