xiaohoua
xiaohoua
您好，请问方便分享一下您使用的数据集吗？万分感谢。[email protected]
Another question, how can i set the number of threads and work units in my filter.
I want to explain what I'm going to do。 I want to keep the text model unchanged and only quantize the visual model of the open_clip model. I'm not sure...
And when i try : ``` import open_clip model, _, transform = open_clip.create_model_and_transforms(model_name="ViT-L-14", pretrained="./test_model/ViT-L -14_laion400m_kd_ViT-B-16_cc3m_12m_ep32.pt") ``` it get same error. but when i load open_clip model by : `model, _,...
thanks for your quick reply. So you mean that the 2-bit dequantization speed of the gguf format model on xpu is relatively faster now? Can you share a demo or...
OK, so gguf is for cpu. another question, Will you open source your kernel implementation in the next release version?
对比3090-4卡和910B4卡训练输出发现：二者输出逻辑不一样：npu4张卡会输出4遍同样的数据，3090则是分开输出。 910B输出：  3090输出： 
Hi. do we have a plan to fix it? now if we install torch =2.8.0+xpu or torch=2.9.0+xpu, it will Automatically install pytorch-triton-xpu, if i use auto-round, get WARNING: ``` 2025-11-26...