JohnnyRacer

I believe this is an issue with the model not being able to align the face properly on close up shots of faces due to a lack of detail for...

The projection results you get by directly projecting to the latent space is usually of poor quality and of low similarity to the source. If your use case is able...

DId you extract the W, S, S_mean, S_std using `extract.py ` first? The manipulator needs to load in these weights to perform the edits.

@ClashSAN You are most likely using fp16 on Pascal based cards (10xx series GPUs) due to the lowmem flag, this is a problem since the support for fp16 compute is...

@17314642 Are the weights loaded in FP16 when using the GPU? There's a bug where the FP16 generation is bugged, on the CPU it's probably forced to use FP32 so...

I can also confirm that finetuning a Seq2SeqLM model works fine, and there are no problems when I test PyTorch functions that uses CUBLAS such as the following snippet: ```python...

> > Hi @JohnnyRacer Thanks for your interest in this! Would be able to share with us the full notebook so that we can try to reproduce? Also can you...

@danielhanchen Sorry I don't really follow what you mean since I already specified `dataset_text_field="text"` in the args when I inited the `SFTTrainer` instance. If you don't mind, can you clarify...

@danielhanchen I think I have solved it, if I add `packing=False` to the training args the `SFTTrainer` seems to initialize and train fine.

@OneCodeToRuleThemAll I don't actually remember the exact dataset that worked since I was just testing a bunch of my own. I think its [this one](https://huggingface.co/datasets/HuggingFaceH4/deita-10k-v0-sft) that worked. It seems like...