lixiaolong
b386f003a0
- 集成Qwen3.8-27B-FP8快速模型和Qwen3.8-27B精确模型作为SGLang服务 - 添加SGLang API配置选项(SGLANG_FAST_BASE_URL、SGLANG_ACCURATE_BASE_URL等) - 实现音频分类中的决策聚合算法(topK、nearestWeight、labelMaxDistance) - 添加视频采样帧限制(VIDEO_SAMPLING_FRAME_LIMIT)和上下文token限制 - 更新健康检查以监控SGLang服务状态 - 实现视频分析的双模式决策策略(快速+精确) - 添加音频参考文件导入工具(import_audio_references.py) - 扩展音频分类标签支持FIND_VEHICLE_HORN类别 - 优化视频分析的帧采样策略，始终包含视频尾部帧 - 添加决策策略参数(tuning、decisionPolicy)支持 - 更新配置类以支持新的SGLang和视频参数 - 修改compose配置以支持Qwen3.8模型部署 - 更新音频分类测试用例验证聚合逻辑 - 重构视频测试以支持SGLang API格式和决策策略
36 lines
880 B
JSON
36 lines
880 B
JSON
{
|
|
"code": "aima.power_video.v1",
|
|
"name": "爱玛测试视频分析",
|
|
"analysisType": "VIDEO_ANALYSIS",
|
|
"model": "qwen3-vl:32b",
|
|
"promptFile": "prompt.txt",
|
|
"think": false,
|
|
"temperature": 0.1,
|
|
"strategies": {
|
|
"fast": {
|
|
"provider": "sglang",
|
|
"model": "Qwen/Qwen3.8-27B-FP8",
|
|
"fallbackModel": "qwen3-vl:8b-instruct",
|
|
"sampleFps": 1.0,
|
|
"maxWidth": 896,
|
|
"maxOutputTokens": 512,
|
|
"numCtx": 16384,
|
|
"emptyResponseRetries": 0,
|
|
"maxRetryOutputTokens": 4096,
|
|
"keepAlive": "30m"
|
|
},
|
|
"accurate": {
|
|
"provider": "sglang",
|
|
"model": "Qwen/Qwen3.8-27B",
|
|
"fallbackModel": "qwen3-vl:32b",
|
|
"sampleFps": 3.0,
|
|
"maxWidth": 1120,
|
|
"maxOutputTokens": 768,
|
|
"numCtx": 32768,
|
|
"emptyResponseRetries": 2,
|
|
"maxRetryOutputTokens": 8192,
|
|
"keepAlive": "30m"
|
|
}
|
|
}
|
|
}
|