Conversation
There was a problem hiding this comment.
Code Review
This pull request updates the default model from Qwen3.5-27B to Qwen3.6-35B-A3B across documentation, configuration files, and examples. It also renames the public encode method in the Template class to a private _encode method, updating internal calls and documentation accordingly. The review feedback highlights that renaming encode is a significant breaking change for users and custom subclasses, recommending that the method remain public to maintain API stability and improve ergonomics in example code.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request updates the Twinkle framework to support Qwen3.6-35B-A3B, migrating model IDs across documentation, examples, and configurations. It significantly expands the documentation with new guides for components like Gym, Hub, and various Loss/Metric types. A key code change refactors the Template class to make encode a convenience wrapper for batch_encode when handling single trajectories, with corresponding updates applied throughout the codebase. Additionally, the requirement to join the Twinkle-Explorers organization for serverless training access has been removed. The review feedback correctly identifies several documentation examples that still reference Qwen3.5 and should be updated for consistency with the Qwen3.6 rollout.
PR type
PR information
Write the detail information belongs to this PR.
Experiment results
Paste your experiment result here(if needed).