Preserve optimized scene rasterization and mesh prefetch source

code/slat_scene_preparation/64/manifest.json

@@ -1,16 +1,30 @@
 {
-  "base_git_revision": "baea78e23ebe182e677df08360246115635d2aec",
+  "base_git_revision": "58a89da9c608ba2f4f76d554624db731359a372e",
   "files": {
-    "preprocess/slat/scene_targets.py": "b79d792b295289dc40e50af026ce010e2ab8a430c261af063269db034efc6723",
+    "preprocess/slat/scene_targets.py": "4e80e8d8861cc15f9d0b9ded275d321d7c2a98982f6c357eda292111b5f889f3",
     "preprocess/slat/publish.py": "fb13b9232b1fc10b1eee980cae94c42fd3d69aec18d16818a74f987ab2f6a1a6",
-    "preprocess/slat/configs/scenes_64.json": "9bbf54138e47b726993460876c42983f2e6c411afeacd8cf3a03af95ab14f91b",
-    "preprocess/slat/README.md": "ac56b9fff6501dae5d230068aaa6fb1ec49d4325800666db45cfb36f5d4e973d",
-    "preprocess/tests/test_slat_aligned_scenes.py": "1d24b13b600580395bf32a360d9116a12ba2941f0af0cc4a2fa51595bd515dbb",
-    "preprocess/tests/test_slat_scene_publication.py": "6c038b1cd17e2bd0abef93696618c7f792969083c968c0111a1fe69e96594be0"
+    "preprocess/slat/configs/scenes_64.json": "2f54c139bc0c72f49b8c3eee98880fb1eb1d0c6fa0d3406b817a7efde199e319",
+    "preprocess/slat/README.md": "dc5adc5eea5310c2266c508133d00a920295bd496f2e55907db5968a8d5e278f",
+    "preprocess/tests/test_slat_aligned_scenes.py": "449b23c4662356aa3489ecd43fdfd658c2f2fd2469b69fcad51ad76ee32d875f",
+    "preprocess/tests/test_slat_scene_publication.py": "6c038b1cd17e2bd0abef93696618c7f792969083c968c0111a1fe69e96594be0",
+    "preprocess/slat/encode.py": "95321d168b498e338d4525cba2923589af298ca0caba25faa3b8474aab65dc2f",
+    "preprocess/tests/test_slat_encode_parallel.py": "d21f2e943668020b8b209c3910578aa9c1d3f90345a9101840522087285b2299",
+    "preprocess/tests/test_slat_native_equivalence.py": "8896f789467e5fed3b321f0045274a5f46fa729c97b1c99eb68a9037a0ca47a1"
   },
   "resolution": 64,
   "command": "CUDA_VISIBLE_DEVICES=1,2 CUDA_DEVICE_ORDER=PCI_BUS_ID .venv/bin/python -m preprocess.slat.scene_targets --resolution 64 --publish",
   "omni6dpose_command": "CUDA_VISIBLE_DEVICES=1,2 CUDA_DEVICE_ORDER=PCI_BUS_ID .venv/bin/python -m preprocess.slat.scene_targets --resolution 64 --dataset omni6dpose --publish",
   "prerequisites": "Matching PartProbe3D checkout, native encoder environment, model weights, and prepared extent observations with mesh geometry.",
-  "targets": "Scene latents and independent object memberships on native 64-grid reference coordinates."
-}
+  "targets": "Scene latents and independent object memberships on native 64-grid reference coordinates.",
+  "optimized_command": "CUDA_VISIBLE_DEVICES=1,2 CUDA_DEVICE_ORDER=PCI_BUS_ID OMP_NUM_THREADS=1 OPENBLAS_NUM_THREADS=1 .venv/bin/python -m preprocess.slat.scene_targets --resolution 64 --parallel 4 --mesh-workers 12 --raster-workers 8 --views-per-job 16 --prefetch-jobs 4 --publish",
+  "verification": {
+    "cpu_tests": 92,
+    "native_gpu_tests": 2,
+    "native_grids": [
+      32,
+      64
+    ],
+    "benchmark_views": 16,
+    "latent_values_bit_identical": true
+  }
+}

code/slat_scene_preparation/64/source.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9721d8f4edbc163ba04709506ff0792ebd819df11a142003139e96d35dd60a45
-size 33824
+oid sha256:7fef58b18b7108622b7a1e2fba9f184930895699dbbaa846f1ec618c64883509
+size 44265