Implementation:Open compass VLMEvalKit KimiVLAPIWrapper

Field	Value
source	VLMEvalKit
domain	Vision, API_Integration

Overview

KimiVLAPIWrapper provides a VLMEvalKit API adapter for Moonshot's Kimi VL thinking vision-language models.

Description

KimiVLAPIWrapper inherits from BaseAPI and communicates with the Kimi VL API via an OpenAI-compatible chat completions endpoint. It includes a summary extraction function that strips thinking tags from model outputs, supports configurable API base URLs through the KIMI_VL_API_BASE environment variable, and handles image encoding to base64. Authentication uses the KIMI_VL_API_KEY environment variable.

Usage

Use this adapter when evaluating Kimi VL thinking models (such as api-kimi-vl-thinking-2506) through the Moonshot API.

Code Reference

Source: vlmeval/api/kimivl_api.py, Lines: L1-159
Import: from vlmeval.api.kimivl_api import KimiVLAPIWrapper

Signature:

class KimiVLAPIWrapper(BaseAPI):
    def __init__(self, model='api-kimi-vl-thinking-2506', retry=5, key=None,
                 verbose=True, system_prompt=None, temperature=0.8,
                 timeout=360, api_base='OFFICIAL', max_tokens=32768,
                 **kwargs): ...
    def generate_inner(self, inputs, **kwargs): ...

I/O Contract

Direction	Description
Inputs	message — text/image/video content list; model-specific params via kwargs
Outputs	generate() returns str prediction; generate_inner() returns (int, str, str) tuple

Usage Examples

# Example instantiation
model = KimiVLAPIWrapper(model='api-kimi-vl-thinking-2506')
response = model.generate(message)

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment