hardlipay
hardlipay
> 验证是可以修改图片大小,目前我设置为540,但是效果还没评估? 试了finetune训练,直接在源码里改了size,tensor size不匹配,传递不了,应该不可以,要改可能要从头做对齐训练
``` def forward(self, x): x = self.forward_features(x) x = self.forward_head(x) return x ``` [](https://github.com/huggingface/pytorch-image-models/blob/0bc9a63ee5e432c85c8f87d7b116a0430ea3764a/timm/models/eva.py#L579C3-L582C17) just to ``` def forward(self, x): x_h = self.forward_features(x) # right? x = self.forward_head(x_h) return x,x_h...
> # merge代码是你自己写的吗?我咋没在项目看到呢。。 # 以前没用过lora,不知道还要merge。 (没仔细看=、=) merge后就能和原始权重一样用huggingface推理,也可以继续lora训练了是吗?
> I successfully make it running on Win10. The key point is to set the python version to 3.8.1 in order to install several requirements. Then find the right builds...
> Hi, I can reproduce. > > ```python > import torch > import deepspeed > from transformers import AutoModelForCausalLM > from transformers.models.llama import LlamaTokenizer > > tokenizer = LlamaTokenizer.from_pretrained("decapoda-research/llama-7b-hf") >...
> maybe it's opencv version wrong, try `pip install opencv-python==4.5.1.48`. Or you can also reference my Colab version https://github.com/K-tang-mkv/visual-chatgpt-googlecolab yes , i had solve it but,it tips that i have...
> > 请问,支持detection返回bbox吗?很多多模态模型已经支持了,且性能也很优秀,为什么cpmv我测了感觉不支持。 但是cpmv对空间位置关系的感知能力也很强,是什么原因导致放弃了训练这一功能呢? 如果加上detect的预训练,做finetune时有bbox输入对应文本,是否能让模型更容易获取新知识? > > 现在哪些模型支持返回 bbox 了? 很多吧。 最早的工作应该是这个吧 [https://github.com/shikras/shikra](https://github.com/shikras/shikra)
https://github.com/Luodian/Otter/blob/0d12192f665f5e9da1ecb2f23d6a360eb7753771/pipeline/constants.py