Mirrors
/
EveryDream2trainer
mirror of https://github.com/victorchall/EveryDream2trainer.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

update captoin doc again

This commit is contained in:
Victor Hall 2023-06-30 12:41:33 -04:00
parent 4ddaf3349f
commit 7aed4e3e35
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
doc/CAPTION.md
View File

@ -8,12 +8,12 @@ This script uses two example image/caption pairs located in the example folder t
This script currently requires an AMPERE or newer GPU due to using bfloat16.
**Trying out different example image/caption pairs will influence how the system captions the input images.**
**Trying out different example image/caption pairs will influence how the system captions the input images.** Adding more examples slows processing.
Supported models:
* `openflamingo/OpenFlamingo-3B-vitl-mpt1b` Small model, requires 8 GB VRAM a num_beams 3, or 12 GB at num_beams 16
* `openflamingo/OpenFlamingo-9B-vitl-mpt7b` Large model, requires 24 GB VRAM at num_beams 3
* `openflamingo/OpenFlamingo-9B-vitl-mpt7b` Large model, requires 24 GB VRAM at num_beams 3, or 36.7gb at num_beams 32
The small model with more beams (ex. 16) performs well with details and should not be immediately discounted.