{ "baseline_commit": "e6efef8f9a728839dd330ffe4bb00635f33051f5", "candidate_runtime_commit": "1b5af9a", "workload": [ { "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_1.tiff", "coordinates": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/he-1.npz", "count": 64, "anchors": [ [ 9216, 25600 ], [ 17408, 25600 ], [ 9216, 54272 ], [ 17408, 54272 ] ], "size_bytes": 84956350, "source_block_size": [ 4096, 4096 ], "read_level": 0, "tile_size": 224, "selection": "four 4x4 tile grids across separated source TIFF blocks; no tissue filtering" }, { "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "coordinates": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/he-5.npz", "count": 64, "anchors": [ [ 17408, 9216 ], [ 33792, 9216 ], [ 17408, 21504 ], [ 33792, 21504 ] ], "size_bytes": 287303418, "source_block_size": [ 4096, 4096 ], "read_level": 0, "tile_size": 224, "selection": "four 4x4 tile grids across separated source TIFF blocks; no tissue filtering" }, { "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_3.tiff", "coordinates": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/he-3.npz", "count": 64, "anchors": [ [ 13312, 29696 ], [ 29696, 29696 ], [ 13312, 58368 ], [ 29696, 58368 ] ], "size_bytes": 644879226, "source_block_size": [ 4096, 4096 ], "read_level": 0, "tile_size": 224, "selection": "four 4x4 tile grids across separated source TIFF blocks; no tissue filtering" } ], "staging": { "source": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "destination": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "bytes": 287303418, "seconds": 2.1555639379657805, "storage": "tmpfs RAM" }, "correctness": { "full_pipeline_shape": [ 474, 384 ], "storage_outputs_exact": true, "rtol": 0.0001, "atol": 0.0001, "expanded_max_abs_errors": { "expanded-before": 0.0, "expanded-after": 0.00014495849609375, "expanded-ram-after": 0.00014495849609375, "expanded-no-overlap": 0.00014495849609375, "expanded-baseline-recheck": 0.0 }, "prototype_full_pipeline_max_abs_error": 6.29425048828125e-05, "retained_full_pipeline_max_abs_error": 0.0, "retained_full_pipeline_exact": true, "retained_storage_outputs_exact": true }, "runs": { "lunit-before": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.595658992882818, "modes": { "model-only": { "samples_seconds": [ 0.3353297980502248, 0.33865594188682735, 0.3407795208040625, 0.3408910990692675, 0.34249340603128076 ], "median_seconds": 0.3407795208040625, "tiles_per_second": 187.80471270395967, "peak_allocated_bytes": [ 366395392, 366395392, 366395392, 366395392, 366395392 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-before-model-only.pt", "embeddings_sha256": "dd54cebd9337fc70dc713abf91081136aa5f2ecfa78a89d0a5bd4661a35bd63d", "cache_advice": [], "trace_path": "output/gpu-performance-audit/lunit-before-model-only-trace.json" }, "cached": { "samples_seconds": [ 0.5018705090042204, 0.4848254898097366, 0.4776876431424171, 0.47169514116831124, 0.46691507706418633 ], "median_seconds": 0.4776876431424171, "tiles_per_second": 133.9787639868238, "peak_allocated_bytes": [ 385662976, 385662976, 385662976, 385662976, 385662976 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-before-cached.pt", "embeddings_sha256": "2c57e468897914baf797b1e74ff0a90f9811ce57289944f6aff03e33b2bad551", "cache_advice": [], "trace_path": "output/gpu-performance-audit/lunit-before-cached-trace.json" }, "wsi": { "samples_seconds": [ 1.0294036658015102, 1.016378456959501, 1.0092771919444203, 0.9871448820922524, 0.9901522900909185 ], "median_seconds": 1.0092771919444203, "tiles_per_second": 63.41171732683364, "peak_allocated_bytes": [ 385662976, 385662976, 385662976, 385662976, 385662976 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-before-wsi.pt", "embeddings_sha256": "86c4557f82e39cd725f0eb28860edb34ffe83e51c3aa9a3b5f3aedf96d6a17a1", "cache_advice": [], "trace_path": "output/gpu-performance-audit/lunit-before-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "baseline" }, "lunit-after": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "d02b70c737e84e1cf886b64e2d56358c681dfcd53d963f561424bc7e27048b46", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.2652621669694781, "modes": { "model-only": { "samples_seconds": [ 0.33902822504751384, 0.3403479198459536, 0.34470526478253305, 0.34435109701007605, 0.34356568404473364 ], "median_seconds": 0.34356568404473364, "tiles_per_second": 186.28170091535375, "peak_allocated_bytes": [ 366395392, 366395392, 366395392, 366395392, 366395392 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-after-model-only.pt", "embeddings_sha256": "85c2678fcf5b3fc8812f9c488dca436455fe0b91671a0ca5d99ad632d3d18fe0", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 0.9918904495703116, "trace_path": "output/gpu-performance-audit/lunit-after-model-only-trace.json" }, "cached": { "samples_seconds": [ 0.34936067182570696, 0.3498760152142495, 0.35676952195353806, 0.351674861041829, 0.3521293490193784 ], "median_seconds": 0.351674861041829, "tiles_per_second": 181.98628076627773, "peak_allocated_bytes": [ 385662976, 385662976, 385662976, 385662976, 385662976 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-after-cached.pt", "embeddings_sha256": "ed5441917635e7ef1d2914a904a6d5365e7aa90a33c5228c97f3127114ddd16f", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.3583218366171468, "trace_path": "output/gpu-performance-audit/lunit-after-cached-trace.json" }, "wsi": { "samples_seconds": [ 0.8681483869440854, 0.9525006690528244, 0.9705271739512682, 0.9034039420075715, 0.8968644109554589 ], "median_seconds": 0.9034039420075715, "tiles_per_second": 70.84317106008777, "peak_allocated_bytes": [ 385662976, 385662976, 385662976, 385662976, 385662976 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-after-wsi.pt", "embeddings_sha256": "40d54d64385eea132d85aeaccedc123a757cd7198f3bede8a06d597155905d6f", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.1171936993119314, "trace_path": "output/gpu-performance-audit/lunit-after-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "lunit-overlap": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.364802457159385, "modes": { "model-only": { "samples_seconds": [ 0.33647919981740415, 0.338087857933715, 0.3397877309471369, 0.3419679249636829, 0.34290031995624304 ], "median_seconds": 0.3397877309471369, "tiles_per_second": 188.35288673197243, "peak_allocated_bytes": [ 366395392, 366395392, 366395392, 366395392, 366395392 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-overlap-model-only.pt", "embeddings_sha256": "d1079ad67860797e39dafee6f980d549f9a45f8f9e66f092e557837bb77c85ce", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.002918851290366, "trace_path": "output/gpu-performance-audit/lunit-overlap-model-only-trace.json" }, "cached": { "samples_seconds": [ 0.3465514681302011, 0.3474569588433951, 0.34760033898055553, 0.34652135404758155, 0.34922799398191273 ], "median_seconds": 0.3474569588433951, "tiles_per_second": 184.19547621967737, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-overlap-cached.pt", "embeddings_sha256": "8d263dc368d4c294e7c3966e61fa59a1525f4a44e8c1e117d4c185289887649b", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.3748109830136377, "trace_path": "output/gpu-performance-audit/lunit-overlap-cached-trace.json" }, "wsi": { "samples_seconds": [ 0.6847651458811015, 0.6560167910065502, 0.6868594351690263, 0.6462201199028641, 0.6821248598862439 ], "median_seconds": 0.6821248598862439, "tiles_per_second": 93.8244649383884, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-overlap-wsi.pt", "embeddings_sha256": "fb17545d7743ed81a9300a4e1cfb70b24ff4d15808ae2ffef8bf3d59ab1e6998", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.4796076954485058, "trace_path": "output/gpu-performance-audit/lunit-overlap-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "lunit-final": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.7449610461480916, "modes": { "model-only": { "samples_seconds": [ 0.34151207003742456, 0.34382994496263564, 0.3459052359685302, 0.3463978301733732, 0.348128727870062 ], "median_seconds": 0.3459052359685302, "tiles_per_second": 185.02177285868723, "peak_allocated_bytes": [ 366395392, 366395392, 366395392, 366395392, 366395392 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/lunit-final-model-only.pt", "embeddings_sha256": "187699d9e02f63f537cfd0434a3598a5b661468fed952304fef62fbb2b5a57b5", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 0.9851817358297114 }, "cached": { "samples_seconds": [ 0.35082580195739865, 0.34992507682181895, 0.3504042981658131, 0.35021627484820783, 0.3506337769795209 ], "median_seconds": 0.3504042981658131, "tiles_per_second": 182.64616140557408, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/lunit-final-cached.pt", "embeddings_sha256": "e370173f9cf667c5748df36b7c19c0ebe8f43f03fa0a38ffa88e21b7e0c45ce2", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.3632470995443464 }, "wsi": { "samples_seconds": [ 0.6756618320941925, 0.6527289301156998, 0.6396553399972618, 0.6440485899802297, 0.7091126709710807 ], "median_seconds": 0.6527289301156998, "tiles_per_second": 98.04989030998772, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/lunit-final-wsi.pt", "embeddings_sha256": "74fdc53b4ab62e5cf4008bb16eb206064bbe6cde67b71beea67e6e508300c51c", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.5462424681644193 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "phikon-before": { "case": "inference", "parameters": { "model": "phikonv2", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": "2ae989a9c40cffaa27f0a6cb29cc94d1d6f9a5fd", "weight_hub_id": "owkin/phikon-v2" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/phikon.py": "db13486f5ec77d635c3bb51e2f55601c6d41fcdb9e1b4d70da3614f981b95cda" }, "model_load_seconds": 7.933326260885224, "modes": { "model-only": { "samples_seconds": [ 0.7648135269992054, 0.7672604240942746, 0.7712700399570167, 0.773348551010713, 0.774408747907728 ], "median_seconds": 0.7712700399570167, "tiles_per_second": 82.98001566813973, "peak_allocated_bytes": [ 1443831808, 1443831808, 1443831808, 1443831808, 1443831808 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/phikon-before-model-only.pt", "embeddings_sha256": "d814cabb091aa6333e44a83602aa78c939ae3317fa3e03328962c97ad3f9f543", "cache_advice": [], "trace_path": "output/gpu-performance-audit/phikon-before-model-only-trace.json" }, "cached": { "samples_seconds": [ 0.816818559076637, 0.8204871260095388, 0.8179547169711441, 0.8189962450414896, 0.8195874991361052 ], "median_seconds": 0.8189962450414896, "tiles_per_second": 78.14443642138777, "peak_allocated_bytes": [ 1463885824, 1463885824, 1463885824, 1463885824, 1463885824 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/phikon-before-cached.pt", "embeddings_sha256": "de2a7765e2dcc6f65b699309554895e687b7933976d26a2245383b1e5eb4f862", "cache_advice": [], "trace_path": "output/gpu-performance-audit/phikon-before-cached-trace.json" }, "wsi": { "samples_seconds": [ 1.354677168885246, 1.3358711160253733, 1.3340446730144322, 1.3795317029580474, 1.3282938529737294 ], "median_seconds": 1.3358711160253733, "tiles_per_second": 47.90881338195233, "peak_allocated_bytes": [ 1463885824, 1463885824, 1463885824, 1463885824, 1463885824 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/phikon-before-wsi.pt", "embeddings_sha256": "0ecb114837029619d2a161cc3ac6397922ae8ad486cd1fe7475a42d7964c59b4", "cache_advice": [], "trace_path": "output/gpu-performance-audit/phikon-before-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "baseline" }, "phikon-after": { "case": "inference", "parameters": { "model": "phikonv2", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": "2ae989a9c40cffaa27f0a6cb29cc94d1d6f9a5fd", "weight_hub_id": "owkin/phikon-v2" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/phikon.py": "db13486f5ec77d635c3bb51e2f55601c6d41fcdb9e1b4d70da3614f981b95cda" }, "model_load_seconds": 15.070211412152275, "modes": { "model-only": { "samples_seconds": [ 0.767687011975795, 0.7721179809886962, 0.7733295070938766, 0.7769689899869263, 0.7795731369405985 ], "median_seconds": 0.7733295070938766, "tiles_per_second": 82.75903015844818, "peak_allocated_bytes": [ 1443831808, 1443831808, 1443831808, 1443831808, 1443831808 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/phikon-after-model-only.pt", "embeddings_sha256": "42ac34249cb7aa52247ce3b15ff7c80092f60b40f1160e41d7723bf543b096fd", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 0.9973368827673481 }, "cached": { "samples_seconds": [ 0.8192117030266672, 0.8341798409819603, 0.8291205589193851, 0.822437590919435, 0.8220768230967224 ], "median_seconds": 0.822437590919435, "tiles_per_second": 77.81745473045791, "peak_allocated_bytes": [ 1463885824, 1463885824, 1463885824, 1463885824, 1463885824 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/phikon-after-cached.pt", "embeddings_sha256": "a4509403718356c52ffe81c2498dd096badb06780f598cfc4cbfc8d537bec04f", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 0.995815675358299 }, "wsi": { "samples_seconds": [ 1.4101237158756703, 1.3448154160287231, 1.326561186928302, 1.3805247240234166, 1.3734536380507052 ], "median_seconds": 1.3734536380507052, "tiles_per_second": 46.59785975071788, "peak_allocated_bytes": [ 1463885824, 1463885824, 1463885824, 1463885824, 1463885824 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/phikon-after-wsi.pt", "embeddings_sha256": "240c26d43569a6c46442353f05d48b9d8aea0fbfa2251a937a21a0ea1a4bb304", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 0.9726364829622707 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "storage-fresh": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "fresh-reader", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.2751454720273614, "modes": { "wsi": { "samples_seconds": [ 0.6586440629325807, 0.668549319030717, 0.643129751086235, 0.6589046958833933, 0.6506756490562111 ], "median_seconds": 0.6586440629325807, "tiles_per_second": 97.16932650245583, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/storage-fresh-wsi.pt", "embeddings_sha256": "fd88607362c15aebea7a60e5dc66cdfe6af8b92c61a46579c3d642d2ba4e769f", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "storage-advised": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "advised-client-drop", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.8219775510951877, "modes": { "wsi": { "samples_seconds": [ 0.8080929168500006, 0.8744337521493435, 0.8494683799799532, 0.7902971028815955, 0.9439324128907174 ], "median_seconds": 0.8494683799799532, "tiles_per_second": 75.34123871863288, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/storage-advised-wsi.pt", "embeddings_sha256": "d241e22927da1d835716ac3f54fefed4c9cacc7ab674f8e70b731bd06d225e0d", "cache_advice": [ { "before": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } } ] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "storage-ram": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.4546244009397924, "modes": { "wsi": { "samples_seconds": [ 0.641307728132233, 0.6735838330350816, 0.6291685928590596, 0.6255833809264004, 0.7162507819011807 ], "median_seconds": 0.641307728132233, "tiles_per_second": 99.79608414574362, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/storage-ram-wsi.pt", "embeddings_sha256": "a3b874baa8701a904e4384eb6c1ef0e56632b1f2abaeb4f0c04b356b75ca9a5a", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "storage-workers2": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 2, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "62ca92828a09d6f743886366ee7a1558b4b5f502db13d731c0814cafe2b8ff25", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.4939677510410547, "modes": { "wsi": { "samples_seconds": [ 0.5123203240800649, 0.49919081199914217, 0.5019722369033843, 0.5038201950956136, 0.5026777680031955 ], "median_seconds": 0.5026777680031955, "tiles_per_second": 127.31814310035918, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/storage-workers2-wsi.pt", "embeddings_sha256": "d5a5beb8eb6c989d275bd6a8f38adf94414547f54d097a03be19f7a78ee9a89f", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "expanded-before": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.6239124222192913, "modes": { "wsi": { "samples_seconds": [ 5.534224086906761, 5.1117948167957366, 5.311193337896839, 5.127756619127467, 5.305780082941055 ], "median_seconds": 5.305780082941055, "tiles_per_second": 48.24926702542414, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-before-wsi.pt", "embeddings_sha256": "5d219e4addabdacb8970d0d3ec879fbfdbb167496a8fd52606f3481025f948cf", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "baseline" }, "expanded-after": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.5058838890399784, "modes": { "wsi": { "samples_seconds": [ 25.37422814616002, 43.30292222183198, 9.621431984007359, 51.92229465814307, 5.609750916017219 ], "median_seconds": 25.37422814616002, "tiles_per_second": 10.088976836079306, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-after-wsi.pt", "embeddings_sha256": "27fa28893a87fa48067a1165983996ff13239a5351411eda91be312e4b3dc9ff", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 0.2091011420082939 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "he3-before": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_3.tiff", "slide_size_bytes": 644879226, "slide_mtime_ns": 1773937680293422400, "coordinates_sha256": "07a631339d1d9b58e05a24035fbc62003ef0ba9aba1389405da9e55113f89951", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.3300762069411576, "modes": { "wsi": { "samples_seconds": [ 0.9717488191090524, 0.9919911841861904, 0.976258005015552, 1.0230356028769165, 0.876217412063852 ], "median_seconds": 0.976258005015552, "tiles_per_second": 65.55644068596443, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/he3-before-wsi.pt", "embeddings_sha256": "36beb52f27ff88f07026b08812bd01e1d9bd425bddf8d78e33b943717cd9d037", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "baseline" }, "he3-after": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_3.tiff", "slide_size_bytes": 644879226, "slide_mtime_ns": 1773937680293422400, "coordinates_sha256": "07a631339d1d9b58e05a24035fbc62003ef0ba9aba1389405da9e55113f89951", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.4105767558794469, "modes": { "wsi": { "samples_seconds": [ 0.6288426369428635, 0.6037307190708816, 0.598546130117029, 0.6261846059933305, 0.6462487529497594 ], "median_seconds": 0.6261846059933305, "tiles_per_second": 102.20628132254288, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/he3-after-wsi.pt", "embeddings_sha256": "64e9896e6789077735174b4358e13fffbdea5d734415def6808b64d516f6ae8c", "cache_advice": [], "baseline_max_abs_error": 0.00013887882232666016, "speedup": 1.5590578172500622 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "expanded-ram-before": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.348903700010851, "modes": { "wsi": { "samples_seconds": [ 5.157192579936236, 5.256007255986333, 5.114414331968874, 5.044589962810278, 5.0661195169668645 ], "median_seconds": 5.114414331968874, "tiles_per_second": 50.05460711303943, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-ram-before-wsi.pt", "embeddings_sha256": "b90eae467b40bca22bd77f375cf6ace8970e0b3481b13858522a6edb60c6978c", "cache_advice": [], "trace_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-ram-before-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "baseline" }, "expanded-ram-after": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 2.2070600278675556, "modes": { "wsi": { "samples_seconds": [ 16.1516412650235, 9.594718493986875, 3.76928454102017, 4.537781602004543, 3.7665657070465386 ], "median_seconds": 4.537781602004543, "tiles_per_second": 56.415231593982675, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-ram-after-wsi.pt", "embeddings_sha256": "87359eeeaab3aafb32006b5cbaa6dae22204b2cafab3cdb2b991fc94a6e049f2", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.127073707053157, "trace_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-ram-after-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "experiment, not the final runtime" }, "expanded-no-overlap": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.027553060092032, "modes": { "wsi": { "samples_seconds": [ 24.760793576948345, 4.917415538104251, 12.93821953702718, 7.968108821893111, 4.906989936949685 ], "median_seconds": 7.968108821893111, "tiles_per_second": 32.12807527133872, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/expanded-no-overlap-wsi.pt", "embeddings_sha256": "b72ff7e417661997d3beb1befcd9a6d4f1be2b431bca16d01636be439733d1b2", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "diagnostic_override": "Temporary wrapper disabled deferred preload; source hashes do not represent this runtime monkeypatch.", "audit_role": "experiment, not the final runtime" }, "expanded-baseline-recheck": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "65d45cabef9b1d1dcd8b6d5e76b333b05bde3cec96058532f5c91d4af86c0920", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.6474265588913113, "modes": { "wsi": { "samples_seconds": [ 14.151145898038521, 11.809649668168277, 5.04485734202899, 5.105565252015367, 5.12232364108786 ], "median_seconds": 5.12232364108786, "tiles_per_second": 49.97731848619227, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/expanded-baseline-recheck-wsi.pt", "embeddings_sha256": "84bff3ab97ddfe6b43f6f7936f61388ec1f7e374533b159f787ae394a27c1aee", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed." ], "audit_role": "baseline" }, "expanded-telemetry-before": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 2.53713831002824, "modes": { "wsi": { "samples_seconds": [ 5.212460215901956, 5.451267518103123, 5.279726332984865, 5.378661367110908, 5.18878430291079 ], "median_seconds": 5.279726332984865, "sample_resources": [ { "parent_minor_faults": 300578, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 2027244, "full": 2027244 }, "host_memory_psi_us": { "some": 1331541, "full": 1331508 } }, { "parent_minor_faults": 319504, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 2649807, "full": 2649807 }, "host_memory_psi_us": { "some": 1559509, "full": 1559460 } }, { "parent_minor_faults": 289047, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 1141334, "full": 1141334 }, "host_memory_psi_us": { "some": 1228050, "full": 1227904 } }, { "parent_minor_faults": 275472, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 273082, "full": 273077 } }, { "parent_minor_faults": 267812, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 517069, "full": 517059 } } ], "tiles_per_second": 48.48736162718339, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-telemetry-before-wsi.pt", "embeddings_sha256": "435852b5ca769e60b0566386ded63546ae51078b331bd4d9d987185d3f222876", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "baseline" }, "expanded-telemetry-after": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.1601676321588457, "modes": { "wsi": { "samples_seconds": [ 43.96727145696059, 3.7959598819725215, 3.6983962210360914, 3.785355884116143, 14.245272099040449 ], "median_seconds": 3.7959598819725215, "sample_resources": [ { "parent_minor_faults": 392854, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 36784132, "full": 36784132 }, "host_memory_psi_us": { "some": 9026470, "full": 9025666 } }, { "parent_minor_faults": 393262, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 25006, "full": 25005 }, "host_memory_psi_us": { "some": 23743, "full": 23736 } }, { "parent_minor_faults": 392367, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 257971, "full": 257958 } }, { "parent_minor_faults": 395244, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 14341, "full": 14342 }, "host_memory_psi_us": { "some": 1352815, "full": 1352742 } }, { "parent_minor_faults": 406978, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 9588818, "full": 9588818 }, "host_memory_psi_us": { "some": 5023193, "full": 5022653 } } ], "tiles_per_second": 67.44012264612579, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/expanded-telemetry-after-wsi.pt", "embeddings_sha256": "3ffff20ce06cf4927841a1c4644304493c6a1942894af69117b86e4bce54656e", "cache_advice": [], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.3908804352909343 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "experiment, not the final runtime" }, "advised-paired-before": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "advised-client-drop", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "274ad5e354373fc29e1067c8c4cb0afcf23f1130469c29e4dd43b03cb1d6e59d", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.1722737839445472, "modes": { "wsi": { "samples_seconds": [ 1.1450463149230927, 1.0826430420856923, 1.3438390309456736, 1.1440667330753058, 1.1179767979774624 ], "median_seconds": 1.1440667330753058, "sample_resources": [ { "parent_minor_faults": 72122, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 293047, "full": 293037 } }, { "parent_minor_faults": 72982, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 179086, "full": 179058 } }, { "parent_minor_faults": 70520, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 317463, "full": 317438 } }, { "parent_minor_faults": 90633, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 265572, "full": 265546 } }, { "parent_minor_faults": 89389, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 249658, "full": 249636 } } ], "tiles_per_second": 55.940792743763254, "peak_allocated_bytes": [ 347127808, 347127808, 347127808, 347127808, 347127808 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/advised-paired-before-wsi.pt", "embeddings_sha256": "ff8925f52fe28e1b85e286234142bb92c3f5e505b0b27423105545e4eff6d5fe", "cache_advice": [ { "before": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } } ] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "baseline" }, "advised-paired-after": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "advised-client-drop", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.6435851040296257, "modes": { "wsi": { "samples_seconds": [ 0.9146360869053751, 0.7893110068980604, 0.7947489488869905, 0.8140610950067639, 0.9866939659696072 ], "median_seconds": 0.8140610950067639, "sample_resources": [ { "parent_minor_faults": 82471, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 82194, "full": 82192 } }, { "parent_minor_faults": 82436, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 153139, "full": 153134 } }, { "parent_minor_faults": 83607, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 178017, "full": 178012 } }, { "parent_minor_faults": 82507, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 182184, "full": 182177 } }, { "parent_minor_faults": 86295, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 197840, "full": 197824 } } ], "tiles_per_second": 78.61817791386804, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/advised-paired-after-wsi.pt", "embeddings_sha256": "12fbd2b27a520495fc6c32ee1fe546c87f836bb70b17693740fd97546acf845f", "cache_advice": [ { "before": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } } ], "baseline_max_abs_error": 0.00014495849609375, "speedup": 1.4053819057226902 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "experiment, not the final runtime" }, "expanded-cpu-transforms": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.6630681499373168, "modes": { "wsi": { "samples_seconds": [ 7.306218969868496, 3.9711724231019616, 9.989085536915809, 5.260265500051901, 4.1157209780067205 ], "median_seconds": 5.260265500051901, "sample_resources": [ { "parent_minor_faults": 281423, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 3339887, "full": 3339887 }, "host_memory_psi_us": { "some": 3793076, "full": 3792895 } }, { "parent_minor_faults": 284674, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 291827, "full": 291801 } }, { "parent_minor_faults": 375186, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 5106661, "full": 5106661 }, "host_memory_psi_us": { "some": 2054484, "full": 2054309 } }, { "parent_minor_faults": 348996, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 797807, "full": 797807 }, "host_memory_psi_us": { "some": 658013, "full": 657969 } }, { "parent_minor_faults": 316485, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 86591, "full": 86591 }, "host_memory_psi_us": { "some": 133225, "full": 133218 } } ], "tiles_per_second": 48.66674505259747, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/expanded-cpu-transforms-wsi.pt", "embeddings_sha256": "dfef0fd0bcf56927144b9471b0642b4170a05e87c18e03af76c045bae45c4e9b", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.0036995913861708 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "diagnostic_override": "Temporary wrapper restored original CPU preprocessing; runtime scheduling remained candidate. Source hashes do not represent this override.", "audit_role": "experiment, not the final runtime" }, "retained-he5": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "3cd78084d2db4697612f7d0779fb70be17c877a2b2c26fad14fb62ecb4c10d50", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 2.506665961118415, "modes": { "model-only": { "samples_seconds": [ 0.3363200491294265, 0.3386502838693559, 0.3420291349757463, 0.34178124205209315, 0.3421166518237442 ], "median_seconds": 0.34178124205209315, "sample_resources": [ { "parent_minor_faults": 67, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 102023, "full": 102018 } }, { "parent_minor_faults": 0, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 86829, "full": 86817 } }, { "parent_minor_faults": 0, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 78185, "full": 78167 } }, { "parent_minor_faults": 436, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 95667, "full": 95659 } }, { "parent_minor_faults": 412, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 86896, "full": 86892 } } ], "tiles_per_second": 187.254278835599, "peak_allocated_bytes": [ 366395392, 366395392, 366395392, 366395392, 366395392 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he5-model-only.pt", "embeddings_sha256": "31e8a861bb8855a8e4b0e6f78bd9a83da63175998eb36af27f9cb1a3aec40b34", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 0.9970691157829019, "trace_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he5-model-only-trace.json" }, "cached": { "samples_seconds": [ 0.3951533939689398, 0.4524159908760339, 0.3744018110446632, 0.3932650580536574, 0.38929152791388333 ], "median_seconds": 0.3932650580536574, "sample_resources": [ { "parent_minor_faults": 24490, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 31876, "full": 31866 } }, { "parent_minor_faults": 32331, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 33524, "full": 33519 } }, { "parent_minor_faults": 15364, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 30473, "full": 30473 } }, { "parent_minor_faults": 17720, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 25558, "full": 25557 } }, { "parent_minor_faults": 20070, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 33556, "full": 33548 } } ], "tiles_per_second": 162.74011303406414, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he5-cached.pt", "embeddings_sha256": "ad61b326a58cd5c23177b95173aa316783a93a98d5b80431474d27a388ed828f", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.2146709537495726, "trace_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he5-cached-trace.json" }, "wsi": { "samples_seconds": [ 0.7997066481038928, 0.8150293971411884, 0.7395877270027995, 0.7734350650571287, 0.7472825499717146 ], "median_seconds": 0.7734350650571287, "sample_resources": [ { "parent_minor_faults": 76254, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 283633, "full": 283613 } }, { "parent_minor_faults": 83313, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 314906, "full": 314869 } }, { "parent_minor_faults": 76519, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 313256, "full": 313245 } }, { "parent_minor_faults": 71004, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 324577, "full": 324566 } }, { "parent_minor_faults": 70255, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 335395, "full": 335382 } } ], "tiles_per_second": 82.74773525463671, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he5-wsi.pt", "embeddings_sha256": "1b582484204a5676c3c4ba83386bd3a5b5023344376289ef969d6d0b12e87aa1", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.3049281543368758, "trace_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he5-wsi-trace.json" } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "retained runtime" }, "retained-he3": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_3.tiff", "slide_size_bytes": 644879226, "slide_mtime_ns": 1773937680293422400, "coordinates_sha256": "07a631339d1d9b58e05a24035fbc62003ef0ba9aba1389405da9e55113f89951", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "3cd78084d2db4697612f7d0779fb70be17c877a2b2c26fad14fb62ecb4c10d50", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.4074434489011765, "modes": { "wsi": { "samples_seconds": [ 0.74149533896707, 0.7731225159950554, 0.742897551972419, 0.7140554189682007, 0.789215117925778 ], "median_seconds": 0.742897551972419, "sample_resources": [ { "parent_minor_faults": 75357, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 138353, "full": 138348 } }, { "parent_minor_faults": 70617, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 80416, "full": 80410 } }, { "parent_minor_faults": 76243, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 85393, "full": 85380 } }, { "parent_minor_faults": 76818, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 62618, "full": 62613 } }, { "parent_minor_faults": 71098, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 93073, "full": 93065 } } ], "tiles_per_second": 86.14915990782008, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-he3-wsi.pt", "embeddings_sha256": "943b60794081d635ca94b8e4f3a40f922d71d0b4ad4ba3bb6dd577bbfea47df3", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.314121984146472 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "retained runtime" }, "retained-advised": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "advised-client-drop", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "3cd78084d2db4697612f7d0779fb70be17c877a2b2c26fad14fb62ecb4c10d50", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.7571627858560532, "modes": { "wsi": { "samples_seconds": [ 0.9194475910626352, 0.9105047900229692, 0.7347492680419236, 1.2427729340270162, 0.8928760769777 ], "median_seconds": 0.9105047900229692, "sample_resources": [ { "parent_minor_faults": 74001, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 115434, "full": 115433 } }, { "parent_minor_faults": 67903, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 100101, "full": 100093 } }, { "parent_minor_faults": 21648, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 71246, "full": 71244 } }, { "parent_minor_faults": 75832, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 71278, "full": 71260 } }, { "parent_minor_faults": 66370, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 89982, "full": 89978 } } ], "tiles_per_second": 70.29067908405564, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-advised-wsi.pt", "embeddings_sha256": "0d22bc954bd6287f3ed80a693caf00c99cf0d47cca95b03ec7e6f0213434e206", "cache_advice": [ { "before": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } }, { "before": { "res": 32673792, "pages": 7977, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" }, "advisory_only": true, "error": null, "after": { "res": 0, "pages": 0, "size": 287303418, "file": "data/histai/wsi/slide_H&E_5.tiff" } } ], "baseline_max_abs_error": 0.0, "speedup": 1.2565191810209417 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "retained runtime" }, "retained-expanded": { "case": "inference", "parameters": { "model": "lunit", "slide": "/tmp/slide2vec-gpu-perf-slides/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1788791667658590806, "coordinates_sha256": "fce08e9e2fd0a0874acc842dc80c8666c870ad5db051b206957df80f2ecd0edb", "num_tiles": 256, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "3cd78084d2db4697612f7d0779fb70be17c877a2b2c26fad14fb62ecb4c10d50", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.2511038139928132, "modes": { "wsi": { "samples_seconds": [ 29.253737705061212, 3.9748765220865607, 3.990106913028285, 3.9109925490338355, 3.9331403409596533 ], "median_seconds": 3.9748765220865607, "sample_resources": [ { "parent_minor_faults": 363666, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 22892874, "full": 22892875 }, "host_memory_psi_us": { "some": 7696136, "full": 7695608 } }, { "parent_minor_faults": 305449, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 43998, "full": 43998 }, "host_memory_psi_us": { "some": 240355, "full": 240346 } }, { "parent_minor_faults": 307392, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 266946, "full": 266931 } }, { "parent_minor_faults": 266073, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 69575, "full": 69575 }, "host_memory_psi_us": { "some": 446594, "full": 446571 } }, { "parent_minor_faults": 299812, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 315353, "full": 315328 } } ], "tiles_per_second": 64.40451636108084, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-expanded-wsi.pt", "embeddings_sha256": "a4f5bffe6bf7d1c899bb283cecfaac62b9bb8199129657832ee0176c3a4c32c9", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.328274300760754 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "retained runtime" }, "retained-workers2": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 2, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "3cd78084d2db4697612f7d0779fb70be17c877a2b2c26fad14fb62ecb4c10d50", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.6501777058001608, "modes": { "wsi": { "samples_seconds": [ 0.5341331248637289, 0.5564586839172989, 0.5365108090918511, 0.5607734008226544, 0.5345520209521055 ], "median_seconds": 0.5365108090918511, "sample_resources": [ { "parent_minor_faults": 10611, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 23781, "full": 23771 } }, { "parent_minor_faults": 5369, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 48657, "full": 48656 } }, { "parent_minor_faults": 3999, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 37679, "full": 37679 } }, { "parent_minor_faults": 8327, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 42618, "full": 42619 } }, { "parent_minor_faults": 10331, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 34098, "full": 34097 } } ], "tiles_per_second": 119.28930212670356, "peak_allocated_bytes": [ 337494016, 337494016, 337494016, 337494016, 337494016 ], "max_abs_error": 0.0, "embeddings_path": "/data/pathology/projects/clement/code/slide2vec/output/gpu-performance-audit/retained-workers2-wsi.pt", "embeddings_sha256": "f459a35a2708be7cd8ce0bac14ce816ed766116e6ef0f693e0f7f08449a4704d", "cache_advice": [] } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "retained runtime" }, "lunit-cpu-transforms": { "case": "inference", "parameters": { "model": "lunit", "slide": "/data/pathology/projects/clement/code/slide2vec/data/histai/wsi/slide_H&E_5.tiff", "slide_size_bytes": 287303418, "slide_mtime_ns": 1773937659973389500, "coordinates_sha256": "841848f671f4fad269742296e20a73a39c1be7831c39f1a0b209c7ae04281a0e", "num_tiles": 64, "tile_size": 224, "batch_size": 16, "workers": 0, "backend": "openslide", "use_supertiles": true, "precision": "fp32", "threads": 1, "cache_policy": "warm", "repeat": 5, "warmup": 1 }, "environment": { "python": "3.11.15", "numpy": "1.26.4", "hs2p": "4.4.3", "transformers": "4.57.6", "timm": "1.0.29", "torch": "2.7.1+cu128", "cuda": "12.8", "device": "cuda", "gpu": "NVIDIA GeForce RTX 2080 Ti", "weight_revision": null, "weight_hub_id": "1aurent/vit_small_patch8_224.lunit_dino" }, "source_sha256": { "scripts/benchmark_inference.py": "38f67d37c288169089b0da734c840337368010c00a606e5820d23106cc2da4c8", "slide2vec/runtime/batching.py": "4dc47d46e0578d7844d06e0434914d1beb1be42db2f380ad7500413cba4d1aaa", "slide2vec/runtime/preprocessing.py": "9a8f5e6a99733ae1454bffa1886645c491401825eaa6be5cd9adfe9e0af75af0", "slide2vec/data/tile_reader.py": "ac4174f63fa09c831418554722d86c9e9ef37dce344ff070a29174204cd17f9a", "slide2vec/encoders/models/lunit.py": "7f1904e29185559fccf07491445e4529f8d76a468d1c463c280ac94c17aaf800" }, "model_load_seconds": 1.8494079099036753, "modes": { "model-only": { "samples_seconds": [ 0.3401592089794576, 0.3397121769376099, 0.3439028700813651, 0.34356969385407865, 0.3456194601021707 ], "median_seconds": 0.34356969385407865, "sample_resources": [ { "parent_minor_faults": 899, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 111, "full": 111 } }, { "parent_minor_faults": 18, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 0, "full": 0 } }, { "parent_minor_faults": 3, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 0, "full": 0 } }, { "parent_minor_faults": 5, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 2095, "full": 2068 } }, { "parent_minor_faults": 5, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 4636, "full": 4635 } } ], "tiles_per_second": 186.2795268175841, "peak_allocated_bytes": [ 366395392, 366395392, 366395392, 366395392, 366395392 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-cpu-transforms-model-only.pt", "embeddings_sha256": "ea6430b378b74617ebfa2a669f05014e29eeac9ef33b38aa5ba20e0488c13dcb", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 0.991878873195372 }, "cached": { "samples_seconds": [ 0.3777345730923116, 0.3753234068863094, 0.3831482019741088, 0.37878579716198146, 0.3948173529934138 ], "median_seconds": 0.37878579716198146, "sample_resources": [ { "parent_minor_faults": 33145, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 718, "full": 712 } }, { "parent_minor_faults": 2922, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 4816, "full": 4814 } }, { "parent_minor_faults": 20647, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 33988, "full": 33987 } }, { "parent_minor_faults": 2613, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 88484, "full": 88461 } }, { "parent_minor_faults": 6760, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 74102, "full": 74100 } } ], "tiles_per_second": 168.96092852349335, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-cpu-transforms-cached.pt", "embeddings_sha256": "0ba138b0be37db5ad828821a970a4ce0758faaa41be03f38e65576b9592e0d96", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.2611023082740926 }, "wsi": { "samples_seconds": [ 0.7458464070223272, 0.7732273759320378, 0.7314755360130221, 0.733708273852244, 0.7509720488451421 ], "median_seconds": 0.7458464070223272, "sample_resources": [ { "parent_minor_faults": 71348, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 246264, "full": 246252 } }, { "parent_minor_faults": 70407, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 274907, "full": 274899 } }, { "parent_minor_faults": 66166, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 242693, "full": 242683 } }, { "parent_minor_faults": 65575, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 298370, "full": 298366 } }, { "parent_minor_faults": 66646, "parent_major_faults": 0, "cgroup_memory_psi_us": { "some": 0, "full": 0 }, "host_memory_psi_us": { "some": 346616, "full": 346610 } } ], "tiles_per_second": 85.8085517305229, "peak_allocated_bytes": [ 376029184, 376029184, 376029184, 376029184, 376029184 ], "max_abs_error": 0.0, "embeddings_path": "output/gpu-performance-audit/lunit-cpu-transforms-wsi.pt", "embeddings_sha256": "c939081b567ddfe03945c0bf3834ec4a11c4672d50aa03c10075bd368670e5c1", "cache_advice": [], "baseline_max_abs_error": 0.0, "speedup": 1.35319709586562 } }, "limitations": [ "Fixed level-0 coordinates; excludes tissue detection, tiling and output persistence.", "Model-only includes CPU output transfer; preloaded tensors remain allocated across modes.", "Cache advice is not proof of eviction or server-cold storage; tmpfs is RAM-backed.", "Resource snapshots bracket each timed sample outside its timer. Faults cover only the parent process; host/cgroup PSI microseconds are shared pressure, not necessarily caused by this process. Unavailable counters are null." ], "audit_role": "experiment, not the final runtime", "diagnostic_override": "Temporary wrapper restored original CPU preprocessing; source hashes do not represent this override." } }, "cached_weight_files": { "lunit": { "revision": "c23bc01cfbe022bc1f2a8a9306125a07818e4eb4", "sha256": "0b7e0236ce430064b31485bbb6f85e661dc5eca0cde709e8f04d59f1c2b23e0b", "size_bytes": 86694768 }, "phikonv2": { "revision": "2ae989a9c40cffaa27f0a6cb29cc94d1d6f9a5fd", "sha256": "261ae680fa699b3b951597fd57aa19c02ef735805acb104b93af69b36d928569", "size_bytes": 1213455560 } }, "diagnostic_limits": [ "Expanded candidate runs showed intermittent read-side stalls, including from RAM.", "Native sample collection was stopped after 904 samples; 478 included the WSI reader, with sampling lag/errors.", "Whole-run profiler diagnostic was terminated during an unusually long stall and is not a throughput measurement." ], "memory_pressure_observations": { "cgroup_full_avg10_percent": 62.6, "cgroup_full_avg60_percent": 43.4, "host_full_avg10_percent": 31.02, "note": "Read during anomalous runs; not per-sample attribution." }, "retained_change": "CUDA scheduling overlap; original CPU preprocessing retained. GPU preprocessing prototype rejected.", "qa_retained": { "accounted_embedding_seconds": 9.3819, "data_pipeline_fraction": 0.9607, "data_pipeline_seconds": 9.0132, "embedding_seconds": 11.2501, "end_to_end_seconds": 16.9884, "failed_slides": 0, "forward_fraction": 0.0393, "forward_seconds": 0.3687, "gpu_busy_fraction": 0.0988, "loader_wait_fraction": 0.9068, "max_loader_wait_ms": 1226.718, "mean_forward_ms": 12.2893, "mean_loader_wait_ms": 283.4932, "mean_preprocess_ms": 16.8516, "mean_reader_open_ms": 8.219, "mean_reader_read_ms": 274.4811, "mean_ready_wait_ms": 0.0944, "mean_worker_batch_ms": 282.7275, "slide_artifacts": 0, "slides_total": 1, "slides_with_tiles": 1, "success": true, "tile_artifacts": 1, "tiles_per_second": 27.9015, "tiling_seconds": 5.5551, "timed_batches": 30, "total_tiles": 474 } }