joint text encoder requires 11GB of vram

when I run any model it always loads "joint text encoder" (i'm new to flux and don't know what that is). This thing requires 11GB of vram and always maxes me out since I have 12 GB.

Arch linux
AMD RX 6700XT
Rocm 6.1
python 3.10.4
latest forge (Feb 5)

[Unload] Trying to free 7928.00 MB for cuda:0 with 0 models keep loaded ...
[Memory Management] Current Free GPU Memory: 12068.00 MB
[Memory Management] Required Model Memory: 5154.62 MB
[Memory Management] Required Inference Memory: 1227.00 MB
[Memory Management] Estimated Remaining GPU Memory: 5686.38 MB
Moving model(s) has taken 3.44 seconds
Distilled CFG Scale: 3.5
To load target model KModel
Begin to load 1 model
[Unload] Trying to free 15985.96 MB for cuda:0 with 0 models keep loaded ...
[Unload] Current free memory is 666.91 MB ... 
[Unload] Unload model JointTextEncoder
[Memory Management] Current Free GPU Memory: 5901.52 MB
[Memory Management] Required Model Memory: 11353.05 MB
[Memory Management] Required Inference Memory: 1227.00 MB
[Memory Management] Estimated Remaining GPU Memory: -6678.53 MB

Reproduce the report on Arch Linux with an AMD RX 6700 XT, ROCm 6.1, Python 3.10.4, and the February 5 Forge build. Start from the logged JointTextEncoder load and memory calculations; done means determining why it loads for every model and documenting or implementing a supported way to run within 12 GB of VRAM.