Batch inference support

[quetzalsly]

RTX PRO 6000
59gb vram usage per inference call
90 seconds per output image

python3 '/root/instance/JoyAI-Image/inference.py' --ckpt-root '/root/instance/models/JoyAI_Image_Edit' --prompt 'Move the camera.
- Camera rotation: Yaw 180°, Pitch 30°.
- Camera zoom: unchanged.
- Keep the 3D scene static; only change the viewpoint.' --image '/root/instance/in/kik_test.png' --output '/root/instance/out/kik/a.png' --seed '8954' --steps '50' --guidance-scale '4' --neg-prompt '' --basesize '1024' --hsdp-shard-dim '1'
- ```

Currently the model loads and unloads to vram on every call.
I would like to ask if its possible to implement batch mode.

--prompts_file '/root/instance/in/prompts.json'

Maybe some other tricks to speed it up?

Thanks.