Pin torch==2.10.0/torchvision==0.25.0 for ZeroGPU

Pin torch==2.4.0/torchvision==0.19.0 for ZeroGPU compatibility

Pin fastapi==0.112.2 to fix TemplateResponse crash on gradio 4.44.0

Pin python_version to 3.10 to fix audioop removal on Python 3.13

Fix runtime crash: pin huggingface_hub==0.25.2 for gradio 4.44.0 compatibility

increase GPU duration

move cached feature to CUDA

doublecheck inference_state["device"]

quick fix in init_state

set device in inference_state manually

make predictor global and remove bf16

autocast

rollback sam2_base and sam2_video_predictor

move segment_with_points to CPU

carefully move between CPU and GPU

upload examples

trim examples

offload_state_to_cpu

change to decorator

remove demo.queue()

build_ext

remove torch.cuda.is_available

fix inference_state["device"]

remove predictor from state

state init

move inference_state to gr.state

session_hash

get session_id

typo

make gr.State individual

move inference_states out of gr.State

move model to cpu when not using ZeroGPU

Initial commit

initial commit