Pin batch size and GPU duration to measured cost

Larger batches with OOM fallback; size GPU duration from measurements

Cast the whole model (incl. fp32 value heads) to bf16

Fix Progress arg splicing; route examples through a 2-arg wrapper

Delete files examples/put_the_box_in_the_drawer_and_close_it.mp4 examples/so101_pink_lego_brick_into_box.mp4 with huggingface_hub

RynnValue-4B demo

document the single-pass inference protocol

add ZeroGPU duration headroom

guard the attention memory budget, default to the reference 64 frames

keep video decode + trend rendering off the GPU; tighten ZeroGPU duration

single-pass per-frame temporal-distance readout

debug: dual readout + gradio6 theme fix

Upload app.py with huggingface_hub

RynnValue-4B temporal distance demo

Upload app.py with huggingface_hub

Upload README.md with huggingface_hub

Upload folder using huggingface_hub

Upload examples/put_the_box_in_the_drawer_and_close_it.mp4 with huggingface_hub

initial commit