araleza

Output images have green tint compared to inputs

1

Hey, thanks for providing this super resolution network, it's produced some great output for me. I do have an issue to report though. As well as upsampling the image, the...

webui install does not appear to install yaml

18

### Describe the bug Hi, I 'git clone'd a fresh checkout of webui, and then ran ./start_linux.sh. It installed some stuff, but then failed with: ``` ******************************************************************* * WARNING: You...

[Feature request] Support GPTQ quantization

35

So I have a GPTQ llama model I downloaded (from TheBloke), and it's already 4 bit quantized. I have to pass in False for the load_in_4bit parameter of: ``` model,...

What's your opinion on DoRA support in unsloth?

1

Looking at the DoRA paper, it seemed to get impressive results:  From: https://arxiv.org/pdf/2402.09353 There even seemed to be some indications that QDoRA could outperform full fine tuning:  Is...

Cannot restart training after training tenc 2 AND using fused_backward_pass

14

If you finetune SDXL base with: ``` --train_text_encoder --learning_rate_te1 1e-10 --learning_rate_te2 1e-10 --fused_backward_pass ``` Then it will train fine. But if you stop training and restart by training from the...

Added --blocks_to_swap_while_sampling, may allow faster sample image generation

1

I've recently switched over to doing Flux full fine tuning instead of LoRA training. But I've found that sample image generation while training is very slow. I'm using `--blocks_to_swap 35`...

Feature to improve training quality via detection of out-of-tolerance latent mean/std values

1

This is a new feature, currently only for Flux LoRA training (although it could be applied to full fine-tune later too at least). It analyses the latents of training images,...

'timestep_sampling=flux_shift' can offer quality improvements, but barely mentioned in docs

6

So I just discovered this Flux (SD3 branch) parameter: ``` --timestep_sampling flux_shift ``` Previously I'd been using ``` --timestep_sampling shift ``` due to: 1) The README.md having `--timestep_sampling shift` in...

Flux: big jumps in key size seems to be due to min_snr_gamma not being hooked up

5

While measuring sd-scripts' LoRA key lengths with Tensorboard, I noticed that from time to time there were big jumps. The jumps seem to correspond to the noise timestep value for...

### Name and Version I'm using the current latest llama-server. ### Operating systems Linux ### Which llama.cpp modules do you know to be affected? llama-server ### Command line ```shell llama.cpp/build/bin/llama-server...