Implementation:Open compass VLMEvalKit JTVLChatWrapper Mini

Field	Value
source	VLMEvalKit
domain	Vision, API_Integration

Overview

JTVLChatWrapper (Mini variant) provides a VLMEvalKit API adapter for the JT-VL-Chat Mini and 2B vision-language models from China Mobile.

Description

This JTVLChatWrapper variant inherits from BaseAPI and communicates with dedicated Jiutian platform endpoints for the mini and 2B model sizes. It selects between two API endpoints based on the model name (jt-vl-chat-mini vs. the 2B variant), supports custom prompt construction for MCQ datasets, and handles image dumping and base64 encoding. Authentication uses a JWT-based app code.

Usage

Use this adapter when evaluating JT-VL-Chat Mini or 2B models through the China Mobile Jiutian AI platform API.

Code Reference

Source: vlmeval/api/jt_vl_chat_mini.py, Lines: L1-284
Import: from vlmeval.api.jt_vl_chat_mini import JTVLChatWrapper

Signature:

class JTVLChatWrapper(BaseAPI):
    def __init__(self, model='jt-vl-chat-mini', retry=5, wait=5, api_base='',
                 app_code='', verbose=True, system_prompt=None,
                 temperature=0.7, max_tokens=2048, proxy=None,
                 **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 = JTVLChatWrapper(model='jt-vl-chat-mini')
response = model.generate(message)

Related Pages

Page Connections

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

Principle

Implementation

Heuristic

Environment