Sana

Sana copied to clipboard

[BUG] FSDP error: Could not find the transformer layer class SanaBlock in the model

8

``` 25-01-05 17:35:47 - [Sana] - INFO - Load checkpoint from /home/linjl/zzc/Sana_1600M_1024px_MultiLing.pth. Load ema: False. 2025-01-05 17:35:47 - [Sana] - WARNING - Missing keys: ['pos_embed'] [rank0]: Traceback (most recent call...

Pevernow

Controlnet Support

5

Thanks for your work, are you gonna train Control Nets for this model?

Oguzhanercan

Can inference be done at FP8? for both 1K and 2K models

9

People asking me to further reduce VRAM usage. Currently 1K model uses 8.7 GB minimum with VAE offloading. If we could do inference at FP8 that would reduce VRAM usage...

FurkanGozukara

Info about other tokenizer

8

Amazing work. An info about the tokenizer, in the tokenizer and embedded code there are references about Qwen2-0.5B-Instruct and 1.5B; I wanted to know if there have been any tests...

chri002

Problem launching training with less GPUs

10

Hi. I setup a Sana training session with one 4090 GPU on a PC, everything was fine so I moved the config and the checkpoint to a PC with 7...

AfterHAL

Adding temp code to run SANA training with litdata

1

midhunharikumar

Using tar shards to train sana

3

Hi, Reading the readme i see that the dataformat seems to look like a uncompressed webdataset. Is it possible to keep the tars uncompressed? Otherwise it's hard for inode usage....

nicolas-dufour

Question: How to construct wids-meta.json for SanaWebDataset?

6

Could you please tell me how to use non-square images for training in this project or the original project? The official example only has the ImgDataset type, but according to...

Pevernow

VAE caching & bucketing

9

Implemented caching for VAE embeddings and local bucketing support. I decided not to implement caching for text embeddings because they consume an excessive amount of disk space, and the text...

Muinez

clip

4

Could you please try training Sana together with CLIP, similar to how it's done in SDXL? I experimented with fine-tuning Sana on CLIP embeddings (I modified the caption channels), and...

Muinez