Owner

thuml

2 indexed repositories · View on GitHub

Transolver-3

Python · 34 stars
Brep2Shape

Python · 11 stars

Beginner friendly Everything

Open All

Mixed-precision training workflow for DrivAerML? Open

Difficulty 3/5 1-2 days Newbie friendliness 52/100

thuml/Transolver-3#6 · Opened Aug 14, 2026
Eval mode 2 does part-wise Rel-L2 averaging, not joint full-mesh Rel-L2 Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

thuml/Transolver-3#5 · Opened Aug 4, 2026
Clarify train loss vs reported Rel-L2 (NASA-CRM / AhmedML / DrivAerML) Open

Difficulty 3/5 1-2 days Newbie friendliness 48/100

thuml/Transolver-3#4 · Opened Aug 1, 2026
Request for exact AhmedML and DrivAerML train/validation/test splits Open

Difficulty 3/5 1-2 days Newbie friendliness 45/100

thuml/Transolver-3#3 · 2 comments · Opened Jul 28, 2026
Finetuning loader requires pretraining reconstruction targets (uv_face_points/uv_edge_points) that the CE loss never uses Open

Difficulty 2/5 1-3 hours Newbie friendliness 76/100

thuml/Brep2Shape#7 · Opened Jul 17, 2026
Which chord-to-arc tau was Brep2Shape-250k tokenized with? Paper says 0.995, shipped BRT call sites default to 0.7 Open

Difficulty 2/5 1-3 hours Newbie friendliness 62/100

thuml/Brep2Shape#6 · Opened Jul 12, 2026
Edge/face transformer attention is O(E^2) per part with no cap, OOMs on high-edge-count B-reps Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

thuml/Brep2Shape#3 · Opened Jul 10, 2026
Release Transolver-3 checkpoints on Hugging Face Open

Difficulty 3/5 1-2 days Newbie friendliness 48/100

thuml/Transolver-3#2 · Opened Jun 30, 2026
About DriveML Dataset Open

Difficulty 1/5 Under an hour Newbie friendliness 35/100

thuml/Transolver-3#1 · Opened Jun 29, 2026

thuml

Get new issues in your inbox