-
Notifications
You must be signed in to change notification settings - Fork 133
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
【Hackathon 5th No.83】PaddleMIX ppdiffusers models模块功能升级同步HF -part #322
Conversation
Thanks for your contribution! |
你好,总的来看当前代码没有什么问题。但是当前无法直接合入这里的PR,因为还需要等待我们将当前develop的CI、CE给修复通过后,你这里没有太大的影响后我们才会合入 |
好的 |
这个可否再进一步升级到diffusers最新的0.23版本呢,0.23下会包含当前社区热门的LCM-LoRA等模型,会挺重要的 |
@guoshengCS 麻烦看下还有什么修改,使用网络可以访问的环境测试是可以 |
好的 |
@co63oc 当前已经发布了0.19.4了,你这里重新将develop的内容合入到你当前分支,然后等所有ci通过后就可以合入了 |
已合并develop分支CI完成 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
|
||
# Non-power of 2 images must be float32 | ||
if (W & (W - 1)) != 0 or (H & (H - 1)) != 0: | ||
x = x.to(dtype=paddle.float32) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
paddle的tensor没有to方法,要用cast或者astype
x_freq = ifftshift(x_freq, axes=(-2, -1)) | ||
x_filtered = ifftn(x_freq, axes=(-2, -1)).real | ||
|
||
return x_filtered.to(dtype=x_in.dtype) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
paddle的tensor没有to方法,要用cast或者astype
@co63oc
@shiyutang tangshiyu你这里也知悉一下,当前需要回滚一下 |
@co63oc 基于之前的工作我们这边在做ppdiffuers 0.24版本的升级,有拉一个分支出来 https://github.com/PaddlePaddle/PaddleMIX/tree/upgrade_ppdiffusers0240 ,也会在快乐开源放一些新题目出来,请问您那边有兴趣一起不 |
好的参加任务 |
赞,我们后面会同步出来一些任务,感谢支持和贡献,一起共建 |
@co63oc 关于fd推理的问题不需要再看了,你这里需要对比一下旧版0.19.4的代码跟你当前升级后的代码里面recompute粒度在哪里发生了变化 |
@JunnYu 现在问题是 |
|
…ddle#322) PaddleMIX ppdiffusers models模块功能升级同步HF https://github.com/PaddlePaddle/community/blob/master/hackathon/hackathon_5th/%E3%80%90PaddlePaddle%20Hackathon%205th%E3%80%91%E5%BC%80%E6%BA%90%E8%B4%A1%E7%8C%AE%E4%B8%AA%E4%BA%BA%E6%8C%91%E6%88%98%E8%B5%9B%E5%A5%97%E4%BB%B6%E5%BC%80%E5%8F%91%E4%BB%BB%E5%8A%A1%E5%90%88%E9%9B%86.md#no83paddlemix-ppdiffusers-models%E6%A8%A1%E5%9D%97%E5%8A%9F%E8%83%BD%E5%8D%87%E7%BA%A7%E5%90%8C%E6%AD%A5hf PaddlePaddle#259 PaddlePaddle#260 PaddlePaddle#262 PaddlePaddle#264 PaddlePaddle#265 升级 models 增加scale 增加get_processor diffusers上LoRA类逐渐取消,改为使用lora_layer 部分测试误差增加,测试修改前代码同样有误差 增加tests/lora目录 flax模型paddle没有使用,没有迁移 test_attention_processor.py ![image](https://github.com/PaddlePaddle/PaddleMIX/assets/4617245/3a7cdfcf-6313-42f7-8bb3-9fd7d46b36cf) test_models_unet_2d.py ![image](https://github.com/PaddlePaddle/PaddleMIX/assets/4617245/51f9ec37-e0f2-4010-b621-966d596bf451) test_lora_layers.py ![image](https://github.com/PaddlePaddle/PaddleMIX/assets/4617245/d4761fde-1eda-4329-8f94-79d4ed6535b3) test_models_unet_motion.py ![image](https://github.com/PaddlePaddle/PaddleMIX/assets/4617245/f3ff8f8c-89f2-46e2-a733-4b6c442e36b5)
…addlePaddle#354) Reverts PaddlePaddle#322 回滚该PR,等之后修复后再合入,下面是对应的问题。 1. fastdeploy的sd和sdxl 推理造成了错误。无法顺利加载vae的decoder模型,可能当前升级有部分修改,导致这个发生错误。 2. sd的预训练和lora微调存在性能下降10%,可能怀疑是recompute粒度发生了修改,部分模块之前没有加recompute现在加上了recompute导致的。
PaddleMIX ppdiffusers models模块功能升级同步HF
https://github.com/PaddlePaddle/community/blob/master/hackathon/hackathon_5th/%E3%80%90PaddlePaddle%20Hackathon%205th%E3%80%91%E5%BC%80%E6%BA%90%E8%B4%A1%E7%8C%AE%E4%B8%AA%E4%BA%BA%E6%8C%91%E6%88%98%E8%B5%9B%E5%A5%97%E4%BB%B6%E5%BC%80%E5%8F%91%E4%BB%BB%E5%8A%A1%E5%90%88%E9%9B%86.md#no83paddlemix-ppdiffusers-models%E6%A8%A1%E5%9D%97%E5%8A%9F%E8%83%BD%E5%8D%87%E7%BA%A7%E5%90%8C%E6%AD%A5hf
#259
#260
#262
#264
#265
#266
#263
#261
升级 models
增加scale
增加get_processor
diffusers上LoRA类逐渐取消,改为使用lora_layer
部分测试误差增加,测试修改前代码同样有误差
增加tests/lora目录
flax模型paddle没有使用,没有迁移
test_attention_processor.py
test_models_unet_2d.py
test_lora_layers.py
test_models_unet_motion.py