王召德
王召德
`cv` functions are based `Express::VARP`, not `Tensor`, so `imread` return type is `VARP`， if you want get tensor, can using below code: ``` MNN::Express::VARP input_var = MNN::CV::imread("0.jpg"); MNN::Tensor* input_tensor =...
Using `Express::_Pad` to pad VARP https://github.com/alibaba/MNN/blob/b2fb0714fc7016f2a927acf487d3e7dc20fe52c0/express/NeuralNetWorkOp.cpp#L731
没有对MIPS平台进行过编译测试
这个是已知问题，代码已经修复了 近期会合并同步
Qwen3-Omni is not supported yet. We expect to add support after the smaller parameter version is released.
预计今天会发布模型，代码近期会同步
Thanks for reaching out and suggesting these new models. Here is the current status: * **DeepseekV3**: This model is too large for on-device deployment, so we will not be supporting...
`Qwen3-VL` has been supported on [v0.0.4](https://github.com/wangzhaode/llm-export/releases/tag/v0.0.4). MNN Models: https://huggingface.co/collections/taobao-mnn/qwen3-vl-mnn https://modelscope.cn/collections/Qwen3-VL-MNN-f4da0cedb82847
microsoft_Phi-4-mini-reasoning gemma-3-4B-it-QAT gemma-3-1B-it-QAT 这些模型在支持中
kv部分目前公开的代码只有mmap的简单实现版本，还没有预取的实现。 KV的优化还在改进中，包括计算部分后续还会修改，感谢您的关注！