modelscope / modelscope/DiffSynth-Studio

Qwen-Image-Edit FP8 inference without CPU offloading

Open
#935 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.1k
Forks
1.3k
Avg merge
13h 12m
Merged PRs (30d)
45

Description

Description:
Hi, thanks for the great work!

I’d like to use the FP8 version without offloading to CPU. The CPU→GPU transfer takes too long for my use case.

I’ve tried different combinations based on the example code here:
example code

However, I can’t seem to skip the overhead caused by this line:
link to code

Could you briefly explain how FP8 works during model inference, and whether it’s possible to run FP8 fully on GPU without offloading?

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

Start with examples/qwen_image/model_inference_low_vram/Qwen-Image-Edit.py and diffsynth/pipelines/qwen_image.py around line 426, then trace the FP8 inference and CPU-offloading path. Determine whether the existing implementation supports fully GPU-resident inference and document the behavior; if a code change is needed, define its scope first.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.