support kimi_k25 mm#113
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request adds support for image inputs in the Kimi-K25 model (KimiK25Vit) by implementing image feature extraction, projection, and embedding masking in get_inputs_embeds, and introduces a placeholder PackedQuantizationCompressor class. The review feedback highlights a potential TypeError when passing a list of tensors directly to the projector instead of mapping over them, and recommends replacing a hardcoded image token ID with a dynamic configuration lookup.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request introduces support for packed dequantization of INT4/INT8 weights in the compressed-tensors format by adding a new PackedDequantizer utility. It integrates this dequantizer into the GPT bridge for loading MLP weights and updates the Kimi-VL model to support multi-modal inputs and image feature extraction. The review feedback identifies two critical issues in gpt_bridge.py: a potential TypeError when quantization_config is None, and a logic error where gate_proj.weight_packed is checked instead of down_proj.weight_packed when loading down-projection weights.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request introduces support for compressed-tensors quantization by adding a PackedDequantizer utility and integrating it into the GPT bridge to unpack INT4/INT8 weights. It also enables multimodal input support for the Kimi-K25 model. The review feedback highlights critical issues, including a redundant projection of image features in kimi_vl.py that could lead to incorrect outputs, the need to safely handle quantization_config when it is not a raw dictionary, and missing safety checks and zero-point handling when converting packed weights in gpt_bridge.py.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
2 participants
No description provided.