fix compatibility with peft and diffusers 0.26.1 #626
Conversation
chengzeyi
commented
Feb 5, 2024
•
chengzeyi
commented
- Fix converter warning for LoraCompatibleLinear.
- Fix import error for diffusers==0.26.1 which causes patch failure.
- Change default log level to warning to make mistakes more noticeable.
- Check if libs are available before apply patches.
| return flow.nn.Linear | ||
| # Workaround for Linear and LoRACompatibleLinear. | ||
| if inspect.isclass(obj): | ||
| return obj |
There was a problem hiding this comment.
@strint 不行的，我这里转LoraCompatibleLinear还是会有问题，diffusers==0.25.1也会
不过也不是不能跑。而是会打很多warning出来，因为转失败了
There was a problem hiding this comment.
很多warning出来具体是？ 我用 python examples/text_to_image_sdxl_lora.py diffusers.version='0.25.1' 测试没有很多warning出来
There was a problem hiding this comment.
现在main分支上的没有warning是因为onediff里把log level设置成error了，这反而造成出现很多问题不好定位，比如和新版本的diffusers的兼容性问题，这次也把这个修了
There was a problem hiding this comment.
顺便建议把onediffx的diffusers的版本要求放宽一点，现在diffusers更新的还是比较快的，已经到0.26了
There was a problem hiding this comment.
我也建议 warning 要报出来。
这里我的意思是 153 重复了。 我在现在main分支上测的时候设置了 log level 为 debug ,
这个 inspect.isclass(obj): 和main 下面代码重复（记得是修复了 Unsupported type: <class 'type'> e=TypeError(' 这种警告。）
@torch2oflow.register
def _(mod: type):
if not is_need_mock(mod):
return mod
return proxy_class(mod)There was a problem hiding this comment.
好像是没有修复，我是今天刚拉的？难道是main分支今天有变动？
There was a problem hiding this comment.
嗯是的，今天的改动，我刚刚 Update Branch，可以试试把 if inspect.isclass(obj): 注释掉实验下
There was a problem hiding this comment.
嗯是的，今天的改动，我刚刚 Update Branch，可以试试把 if inspect.isclass(obj): 注释掉实验下
👍刚才试了一下已经修复了，所以已经把这段删掉了
| @@ -37,7 +37,7 @@ def configure_logging(self, name, level, log_dir=None, file_name=None): | |||
|
|
|||
| # Create a console formatter and add it to a console handler | |||
| console_formatter = ColorFormatter( | |||
| fmt="%(levelname)s [%(asctime)s] - %(message)s", datefmt="%Y-%m-%d %H:%M:%S" | |||
| fmt="%(levelname)s [%(asctime)s] %(filename)s:%(lineno)d - %(message)s", datefmt="%Y-%m-%d %H:%M:%S" | |||
There was a problem hiding this comment.
有看打印出来的日志，filename不为none，还是有意义的
|
CI 挂了，不知道是啥原因，神奇 |
|
哦，好像CI挂了和我这个PR没啥关系，所有PR都挂了😄 |
是的，报错是因为 oneflow 那边一个修复还没编译出来 |
|
does this mean we can use peft and multiple LoRAs (for slow inference mode)? |
@isidentical for peft, you can try this example #530 PR 530 is not related to this PR. This is the original version of peft load. @marigoold is working on a fast load on peft. |
|
also before I forget, onediffx is still uninstalled with diffusers==0.26.1 even with this PR: |
@chengzeyi need to make a test with onediffx |
已修改 |
|
@chengzeyi can you do 0.26.2 (latest version)? Instead of 0.26.1 |
Changed but I hope the requirement could be loosened. |
