AlexanderZhx

[BUG] Exporting / downloading model larger, than VRAM available (trained with DeepSpeed) fails

8

I trained a 33b model with DeepSpeed on 40GB cards. Based on the traceback, the model seems to be too large to fit into one GPU. Is it possible to...

**Bug Description** Likely because of this https://stackoverflow.com/questions/45940861/android-8-cleartext-http-traffic-not-permitted **Steps to Reproduce** Use an opena-ai compatible endpoint via http, ip and port on android

[Doc]: Clarify QLoRA (Quantized Model + LoRA) Support in Documentation

7

### 📚 The doc issue Two parts of the documentation appear to contradict each other, especially at first glance. Here, it is explicitly stated that LoRA inference with a quantized...

[Bug]: GPU memory does not get cleared after stopping training. (SD1.5)

4

### What happened? When stopping training in the GUI (Stop Training button) with SD1.5 fine-tuning, GPU memory does not get cleared. So next run OOM-s Tested on `e895ddc`, where it...