publicstaticvo
publicstaticvo
@cmikeh2 this is another error with zero_stage=3
I deleted `free_param(param)` at line 1115 of `deepspeed/runtime/zero/partition_parameters.py` and it seems to work, but I don't know if it is the right solution. For example, I met another Cuda Out...
https://space-mm-data.oss-cn-wulanchabu.aliyuncs.com/downstreamv2/iemocap.tgz https://space-mm-data.oss-cn-wulanchabu.aliyuncs.com/downstreamv2/mintrec.tgz https://space-mm-data.oss-cn-wulanchabu.aliyuncs.com/downstreamv2/mosei.tgz https://space-mm-data.oss-cn-wulanchabu.aliyuncs.com/downstreamv2/mosi.tgz 这些是pre-processed fine-tuning data,都是由pickle组成可以直接用,spokenwoz和预训练的晚一点给
你的transformers版本是4.18么 不过按理来说应该没问题的,因为我这边4.28也能跑通 我看了整个代码都没找到有这个“cache_dir” 你的traceback是不是没有截取完整,我看不出是代码执行到哪出的问题
哦 源代码里面好像是没给直接从一个wavlm和一个roberta直接下游微调的设置,必须要从预训练好的模型开始的 你可以自己试试改改,应该比较简单,参考ATForPretraining的写法就成