Feature request
When using Transformers to run the Xiaomi-MiMo-V2.5 model, I found that Transformers is not very compatible with Xiaomi's FP8 quantization processing and custom MOE model architecture. I hope that Transformers can be made more compatible with similar issues.
model address and instruction：https://modelscope.cn/models/XiaomiMiMo/MiMo-V2.5
Motivation
I want to use the transformers library to get the MIMO model working, which will help me with the SFT framework.
Your contribution
Please leave a message if you need more information.
Feature request
When using Transformers to run the Xiaomi-MiMo-V2.5 model, I found that Transformers is not very compatible with Xiaomi's FP8 quantization processing and custom MOE model architecture. I hope that Transformers can be made more compatible with similar issues.
model address and instruction：https://modelscope.cn/models/XiaomiMiMo/MiMo-V2.5
Motivation
I want to use the transformers library to get the MIMO model working, which will help me with the SFT framework.
Your contribution
Please leave a message if you need more information.