lllyasviel / lllyasviel/stable-diffusion-webui-forge

odd vae decoding behaviour when using fp16 flux +fp16 t5xxl

Open
#1,768 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13k
Forks
1.7k
PR merge metrics
No merged PRs in 30d

Description

when using fp16 flux + fp16 t5xxl, generating works just fine but when it reaches the vae decoding stage, it allocates like 10 gb of my ram, overflowing the ram into the ssd and making the pc super unresponsive just for the vae and i have no idea why, the vae is like 200 mb it should have no problems.

also after the vae has finally decoded all models get unloaded from ram and vram, making it so i have to load the models again for generating another image.

i do not believe this to be a memory problem since i have rtx 4090 and 32 gb of ram, and generating speed is perfectly fine.

with the exact same setup except with nf4 flux instead of fp16 everything works perfectly fine.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No file or test is named. Reproduce with fp16 Flux and fp16 T5XXL, compare the VAE decoding path with the working nf4 Flux setup, and inspect where memory is allocated and models are unloaded. Done means decoding no longer causes excessive RAM or SSD use and models remain available for the next generation.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.