Commit History
-
Bump cadgenbench pin to 8ae1432 (revert 0.2.0 edit-baseline rework; back to 0.1.0)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>dca6ffe
-
Bump cadgenbench pin to 2abaea8 (cadquery optional; fixes libTKIVtk/VTK clash)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>cc78a9a -
Pin VTK stack (pyvista 0.48.4 / vtk-egl 9.3 dev) for reproducible builds
Michael Rabinovich +1: CursorSame drift fix as the leaderboard image: an unpinned pyvista + vtk-egl float to latest on rebuild and mismatch (libvtkWrappingPythonCore*.so not found). Pin pyvista==0.48.4 and vtk-egl to the latest 9.3 cp312 build. Co-authored-by: Cursor <cursoragent@cursor.com>368120e
-
Drop OpenSCAD; pin cadgenbench a47c951 (STEP-only eval)
Michael Rabinovich +1: CursorRemove the openscad apt package (the mesh-submission backend is gone) and bump the pinned cadgenbench commit to a47c951. Co-authored-by: Cursor <cursoragent@cursor.com>84e5b14
-
Bump cadgenbench Docker pin to e7a60e0
Michael Rabinovich41f3c89
-
Install OpenSCAD for baseline jobs
Michael Rabinovich99e7deb -
Bump cadgenbench Docker pin to f284fbd
Michael Rabinovich79200be
-
pin cadgenbench to 11babb1 (baseline robustness fixes)
Michael Rabinoviche831d35 -
pin cadgenbench to 9f1e1a6 (Claude Fable 5 temperature fix)
Michael Rabinovich7ccf2cf
-
Bump CADGENBENCH_SHA to 037eade (GT answer-key edit diff)
Michael Rabinovich +1: CursorPicks up the GT "answer key" edit-diff report column + render helpers so re-evaluated/new submissions generate the blue ground-truth turntable. Co-authored-by: Cursor <cursoragent@cursor.com>682d3b0 -
Bump CADGENBENCH_SHA to ee8b257 (deterministic aligner)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>184438e -
Bump pinned cadgenbench SHA to 0cf5913 (sample-named run_summary keys)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>48f588c -
Rename per_fixture_results report key to per_sample_results
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>891e8a2 -
bump cadgenbench -> 68e9337 (clamp edit-diff zoom to full view)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>64d91b9 -
bump cadgenbench -> 2b1c939 (surface-deviation edit-diff field)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>d84d38f
-
eval-gpu: bump cadgenbench to 193f2f5
Michael Rabinovich +1: CursorPicks up the orientation gizmo on render views + headline-pill polish so eval-generated per-submission reports match the leaderboard. Co-authored-by: Cursor <cursoragent@cursor.com>6c44ae5 -
deploy: bump cadgenbench to b76a55a
Michael Rabinovich +1: CursorPins the eval image to cadgenbench b76a55a so generated reports carry the submission name in the title and deep-link their metric pills to the leaderboard's /metrics explainer. Co-authored-by: Cursor <cursoragent@cursor.com>a48bddb -
eval: pass submission name + metrics URL to report generator
Michael Rabinovich +1: CursorReads submission_name from the unpacked meta.json and forwards it plus the Space's /metrics route to `report single` so the hosted report titles itself with the submission name and its metric pills deep-link to the metrics explainer. Co-authored-by: Cursor <cursoragent@cursor.com>ed16355
-
deploy: bump cadgenbench to 72f6e82 (scale-relative edit-diff tol, viz polish)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>55616a2 -
deploy: bump cadgenbench to 850281d (unified viz colours)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>3bb1b02 -
Pin cadgenbench to 2862586 (sharpen topology scoring + metrics docs)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>1cac53c
-
Pin cadgenbench to d514d6a (report thumbnail grid)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>aeab4c1 -
Pin cadgenbench to 1010043 (report legend fixes)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>969640a -
eval_job: pass submission-zip download URL to the report
Michael Rabinovich +1: CursorBuild the submissions/<id>.zip resolve URL (same blob URL the gallery links) and pass it as --download-url so the hosted report shows a "Download submission ZIP" button. Bump CADGENBENCH_SHA to 0c7690e (report generator with the flag). Co-authored-by: Cursor <cursoragent@cursor.com>1c053c8 -
pin cadgenbench to 10bf32c (HF inference token + GT-optional fixes)
Michael Rabinovichad03e51 -
eval_job: upload interface overlay + link GT/input in hosted report
Michael Rabinovich +1: CursorUpload each fixture's interface_overlay.png to the public render bucket alongside the turntables so every report artifact lands together, and pass --gt-base-url/--input-base-url to `report single` so the hosted report links GT/input/overlay (lazy) instead of base64-inlining them. Bump CADGENBENCH_SHA to 25943a0 (the report generator that understands the new flags). Co-authored-by: Cursor <cursoragent@cursor.com>3cf13d3 -
eval_job: warm the render CDN after upload
Michael Rabinovich +1: CursorAfter uploading a submission's renders to the public bucket, fetch each object's /resolve/ URL once (parallel, anonymous, best-effort) so the first viewer of the gallery/report hits a warm Xet edge cache instead of paying the cold chunk-reconstruction latency. New _warm_render_cdn helper; called in both the single-job and shard paths. _upload_renders_to_bucket now returns the uploaded object paths so they can be warmed. Co-authored-by: Cursor <cursoragent@cursor.com>ce583a4 -
eval_job: upload renders to the public bucket; report references them by URL
Michael Rabinovich +1: CursorThe job is now the sole render uploader: after eval it pushes each fixture's renders to renders/<id>/ in the public HF Storage Bucket (single batch), in both single-job and shard modes. The report is then built with --render-base-url so candidate renders (incl. the edit-diff WebP) are bucket URLs, not base64, and only the report HTML/JSON is committed to the dataset. Dockerfile: bump cadgenbench pin to 3d49822 (--render-base-url support). Co-authored-by: Cursor <cursoragent@cursor.com>2cf3635
-
Pin cadgenbench to a0f7205 (turntable WebP renderer)
Michael Rabinovich +1: CursorBumps CADGENBENCH_SHA so the eval-gpu image installs the cadgenbench build that ships render_mesh_turntable_webp. Co-authored-by: Cursor <cursoragent@cursor.com>e3b5dc2 -
Publish all per-fixture renders (PNG views + rotating.webp)
Michael Rabinovich +1: CursorUpload each fixture's whole renders/ folder under renders/<submission>/<fixture>/ instead of a single iso thumbnail. Co-authored-by: Cursor <cursoragent@cursor.com>e3a820e -
sync eval shards to bucket via API (no mount)
Michael Rabinovich9368361 -
pin cadgenbench to 5f312f3
Michael RabinovichPull the renderer fix that isolates each camera view in its own PyVista plotter so future eval reports do not duplicate the iso image across front/top/right.8032255 -
stage eval shards through mounted bucket
Michael Rabinovich8bfcf51 -
pin cadgenbench to 48d910f
Michael Rabinovichf1dcca9 -
remove temp _probe_job.py
Michael RabinovichOne-off diagnostic probe; no longer referenced.972ff3a -
pin cadgenbench to 724809e
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>0384079 -
pin cadgenbench to 44a3c6b
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>4048b86 -
eval_job: publish report + gallery in one commit
Michael Rabinovich +1: CursorBatch the whole-submission report + per-fixture gallery thumbnails into a single create_commit instead of one upload_file per file, matching the sharded merge path. Faster publish and no 429 "concurrency queue" failures on the dataset commit endpoint. Co-authored-by: Cursor <cursoragent@cursor.com>9b64c98 -
pin cadgenbench to defbc98
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>772cc56 -
pin cadgenbench to 5ed7d1d
Michael Rabinovichf7305d8
-
pin cadgenbench to c92c750
Michael Rabinovichf18d005 -
pin cadgenbench to db51404
Michael Rabinovicha53b52e -
pin cadgenbench to 2d02f83
Michael Rabinovichcf2eb5e -
pin cadgenbench to b3ff8f0
Michael Rabinovichc6245b8 -
dockerfile: allow runtime pip installs in venv
Michael Rabinovichbf4211c -
dockerfile: isolate Python deps in venv
Michael Rabinovichba9a513 -
pin cadgenbench to e446cca (benchmark-v1 aligner + sidecar meshes + Open3D)
Michael Rabinovich +1: CursorCo-authored-by: Cursor <cursoragent@cursor.com>be11b5a -
space: add rebuild and probe helpers
Michael Rabinovich8762996 -
dockerfile: track latest cadgenbench pre-v1
Michael RabinovichInstall cadgenbench from main so the eval image picks up the Open3D alignment dependency before the v1 release pin is set.b1601ec
-
dockerfile: drop github-api cache-bust ADD (now 403s, redundant for pinned SHA)
Michael Rabinovich +1: CursorThe `ADD https://api.github.com/.../commits/${SHA}` line existed only to invalidate the pip layer when a floating ref moved. GitHub's API now returns 403 on the HF build IP, failing the build at that step. With an explicit SHA pinned in the ARG + install URL the pip layer is already deterministic, so the line is removed. Re-pin by bumping CADGENBENCH_SHA. Co-authored-by: Cursor <cursoragent@cursor.com>380ae5d