LIU SIYANG

提示：Scan: Invalid Region ID

尝试提交10000像素的图片到isky,如果开启缩略图功能,服务器的性能不足以支撑转那么大的图的缩略图, CPU和内存基本跑满,然后溢出就会失败, 关闭缩略图后,基本上提交后马上非常快速就返回了外链,但是关闭缩略图后 因为网页默认就会用原图去展示列表,会导致网页展示页列表默认加载10000像素的图,会导致网页非常卡.如果能在使用api提交图片时自定义缩略图,就可以在本地性能不错的机器去完成转换.或者还有其他办法吗?

连续的标点符号时很容易发生参考音频泄露

4

使用分支的apiv2，开启分割5，当片段中存在一串连序的标点符号 比如一连串,,,,,,,,逗号时，这段时间就会疯狂泄露。。试了main中 的webui推理出来的音频没这个问题。

WD14 Used to push prompt words from pictures

Can you execute it?

How to make the progress bar identify the progress of task requests in different queues when requesting api.txt2img asynchronously

1

Instead of just reading the current progress

已经改掉了NCCL, 接着出现: Traceback (most recent call last): File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\train_db.py", line 504, in train(args) File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\train_db.py", line 323, in train encoder_hidden_states = train_util.get_hidden_states( File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\library\train_util.py", line 4215, in get_hidden_states encoder_hidden_states =...

Error during multi-CPU training

1

Traceback (most recent call last): File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\train_db.py", line 529, in Traceback (most recent call last): File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\train_db.py", line 529, in train(args) File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\train_db.py", line 343, in train train(args) File "E:\AI\AIArchitecture\lora-scripts-v1.7.3\sd-scripts\train_db.py",...

[Feature]: Can I add auto-cleanup of video memory when using API requests?

1

### Expected behavior Because it always takes up 3GB of video memory after calling the api, now I will add auto.py ,204 line async def api_category_mask(payload: CategoryMaskRequest = Body(...), .......