SD3: add T5-XXL text encoder support
Open
- Dominant language
- Swift
- Stars
- 2.1k
- Forks
- 202
- Avg merge
- 2d 8h
- Merged PRs (30d)
- 55
Description
SD3 pipeline currently uses dual CLIP (L+G) only, zero-padding the T5 portion of encoder_hidden_states. Adding T5-XXL (256 token seq len) would significantly improve text comprehension for complex prompts.
Contributor guide
No contributing guide indexed for this repository
Research direction
No files or tests are named. Start by locating the SD3 pipeline and the construction of encoder_hidden_states, then trace how the dual CLIP path currently zero-pads the T5 portion. Done means the pipeline incorporates T5-XXL with a 256-token sequence length and supports more complex prompts.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, swift
- Domain
- ai, machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100