{ "device": "NVIDIA GB10", "deployment": { "compose_project": "h3-blackwell-runtime", "service": "h3-hot-runtime", "image": "sha256:b9e2949ad4acbacbda3a16e3e60667f476a6330d5763a2bdfbf9a98150a0babd", "feature_flags": { "H3_NVFP4_MODULATE_FUSION": "1", "H3_NVFP4_SWIGLU_FUSION": "1" }, "state": "running", "health_ready": true }, "startup": { "warmup_request_seconds": 30.482452621974517, "note": "First startup warmup includes one-time CUDA extension compilation." }, "resident_smoke": { "output": "/output/h3-blackwell-runtime/deployed-nvfp4-swiglu-fusion-smoke.mp4", "width": 320, "height": 192, "frames": 22, "steps": 1, "seed": 440420, "attention": "sage2", "text_conditioned_seconds": 1.4687969940132461, "sampled_seconds": 0.2173400520114228, "vae_decoded_seconds": 0.28646593599114567, "request_seconds": 2.172695894027129, "wall_seconds": 2.17405128700193, "media_validation": { "codec": "h264", "width": 320, "height": 192, "frames": 22, "duration_seconds": 0.916667 } }, "tests": { "nvfp4_cuda_tests": "3 passed" }, "result": "pass" }