TomasGuija/kornia-kimi-vl-a3b-instruct-vision

Kimi-VL-A3B-Instruct vision and projector weights for Kornia

This is an unofficial, temporary development checkpoint. It is hosted in a personal repository to support the implementation and review of Kornia issue #3869. It is not an official Kornia model release, and its repository or location may change after the contribution is reviewed.

This repository contains the vision tower and multimodal projector weights from moonshotai/Kimi-VL-A3B-Instruct, converted to Kornia's KimiVLModel state-dictionary layout.

It does not contain the language-model weights or a complete Kimi-VL model.

Checkpoint

• File: model.pt
• Source: moonshotai/Kimi-VL-A3B-Instruct
• Contents: vision tower and multimodal projector only
• Format: PyTorch state dictionary
• Conversion: resizes the original 64 x 64 positional-embedding grid to the current Kornia model's configured grid

License and attribution

The source model is published under the MIT license. See the official model repository for its model card, license, usage information, and limitations.