Rename the models by version; release unimate_uniml3d_f60_v3_preview as the recommended model and unimate_truebones_f60_v2

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LjgwfhefMhyk7qWfFTek8y

.gitattributes

@@ -34,65 +34,65 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 teaser.png filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/669a684a57bbceb71e5e7e0d_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/669a684a57bbceb71e5e7e0d_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/Ant-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/Ostrich-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/Ostrich-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/f56c1adb37124925b275b3e122176e57-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/f56c1adb37124925b275b3e122176e57-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/mixamo-2_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/mixamo-2_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/mixamo-3_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_000000/mixamo-3_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_010000/669a691d57bbceb71e5f922a_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_010000/669a691d57bbceb71e5f922a_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_010000/Ant-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_010000/Ant-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_020000/669a6dd357bbceb71e656c6f_glb-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_020000/669a75d157bbceb71e6fc819_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_020000/669a75d157bbceb71e6fc819_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_020000/Goat-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_020000/Goat-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_030000/669a746c57bbceb71e6dfa5d_fbx-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_030000/669a746c57bbceb71e6dfa5d_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_030000/ab1c110b22244ba2b91e2224711ec7ef-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_030000/ab1c110b22244ba2b91e2224711ec7ef-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_030000/mixamo-2_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_030000/mixamo-2_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_040000/669a816357bbceb71e7ed0e7_gltf-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_040000/669a816357bbceb71e7ed0e7_gltf-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/669a797057bbceb71e747fa4_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/669a797057bbceb71e747fa4_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/669a89c157bbceb71e89b323_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/Cricket-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/Cricket-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/mixamo-2_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_050000/mixamo-2_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_060000/669a6df657bbceb71e659a0b_fbx-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_060000/669a6df657bbceb71e659a0b_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_070000/Crab-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_070000/Crab-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_080000/BrownBear-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_080000/BrownBear-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_080000/Lynx-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_080000/Lynx-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_080000/mixamo-3_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_080000/mixamo-3_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_090000/Scorpion2-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_090000/Scorpion2-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_100000/Elephant-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_100000/Elephant-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_100000/Hound-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_100000/Hound-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/669a751957bbceb71e6ed97e_fbx-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/669a751957bbceb71e6ed97e_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/669a876857bbceb71e86a371_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/669a876857bbceb71e86a371_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/Crocodile-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/Crocodile-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/Scorpion2-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_preview/samples/step_110000/Scorpion2-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/669a684a57bbceb71e5e7e0d_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/669a684a57bbceb71e5e7e0d_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/Ant-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/Ostrich-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/Ostrich-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/f56c1adb37124925b275b3e122176e57-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/f56c1adb37124925b275b3e122176e57-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/mixamo-2_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/mixamo-2_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/mixamo-3_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_000000/mixamo-3_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_010000/669a691d57bbceb71e5f922a_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_010000/669a691d57bbceb71e5f922a_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_010000/Ant-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_010000/Ant-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_020000/669a6dd357bbceb71e656c6f_glb-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_020000/669a75d157bbceb71e6fc819_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_020000/669a75d157bbceb71e6fc819_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_020000/Goat-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_020000/Goat-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_030000/669a746c57bbceb71e6dfa5d_fbx-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_030000/669a746c57bbceb71e6dfa5d_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_030000/ab1c110b22244ba2b91e2224711ec7ef-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_030000/ab1c110b22244ba2b91e2224711ec7ef-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_030000/mixamo-2_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_030000/mixamo-2_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_040000/669a816357bbceb71e7ed0e7_gltf-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_040000/669a816357bbceb71e7ed0e7_gltf-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/669a797057bbceb71e747fa4_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/669a797057bbceb71e747fa4_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/669a89c157bbceb71e89b323_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/Cricket-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/Cricket-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/mixamo-2_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_050000/mixamo-2_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_060000/669a6df657bbceb71e659a0b_fbx-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_060000/669a6df657bbceb71e659a0b_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_070000/Crab-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_070000/Crab-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_080000/BrownBear-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_080000/BrownBear-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_080000/Lynx-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_080000/Lynx-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_080000/mixamo-3_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_080000/mixamo-3_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_090000/Scorpion2-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_090000/Scorpion2-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_100000/Elephant-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_100000/Elephant-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_100000/Hound-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_100000/Hound-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/669a751957bbceb71e6ed97e_fbx-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/669a751957bbceb71e6ed97e_fbx-4_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/669a876857bbceb71e86a371_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/669a876857bbceb71e86a371_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/Crocodile-1_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/Crocodile-1_ric.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/Scorpion2-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v1_preview/samples/step_110000/Scorpion2-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
 unimate_uniml3d_f60_v2/samples/step_000000/669a67ac57bbceb71e5db0b5_fbx-5_fk.mp4 filter=lfs diff=lfs merge=lfs -text
 unimate_uniml3d_f60_v2/samples/step_000000/669a67ac57bbceb71e5db0b5_fbx-5_ric.mp4 filter=lfs diff=lfs merge=lfs -text
 unimate_uniml3d_f60_v2/samples/step_000000/7f0acec2bcf14dd1967fcc1d10ade876-4_fk.mp4 filter=lfs diff=lfs merge=lfs -text
@@ -225,5 +225,6 @@ unimate_uniml3d_f60_v2/samples/step_100000/mixamo-3_fk.mp4 filter=lfs diff=lfs m
 unimate_uniml3d_f60_v2/samples/step_100000/mixamo-3_ric.mp4 filter=lfs diff=lfs merge=lfs -text
 unimate_uniml3d_f60_v2/samples/step_100000/Rhino-0_fk.mp4 filter=lfs diff=lfs merge=lfs -text
 unimate_uniml3d_f60_v2/samples/step_100000/Rhino-0_ric.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_uniml3d_f60_v2_full_cross_attn/samples/**/*.mp4 filter=lfs diff=lfs merge=lfs -text
-unimate_mixamo_f60/samples/**/*.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_uniml3d_f60_v2_cross_attn/samples/**/*.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_mixamo_f60_v2/samples/**/*.mp4 filter=lfs diff=lfs merge=lfs -text
+unimate_truebones_f60_v2/samples/**/*.mp4 filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -6,3 +6,7 @@
 # Training samples: only the videos are uploaded
 */samples/**/*.npy
 */samples/tpose/
+# unimate_uniml3d_f60_v3_preview: training logs stay local
+unimate_uniml3d_f60_v3_preview/logs/
+unimate_uniml3d_f60_v3_preview/slurm_14924666.out
+unimate_uniml3d_f60_v3_preview/slurm_14924666.err

README.md

@@ -23,141 +23,125 @@ tags:
 
 ![UniMate teaser](teaser.png)
 
-Pretrained checkpoints for UniMate, a text-conditioned flow-matching model that generates motion for skeletons of arbitrary topology: animals, humanoids and rigged objects.
+Pretrained checkpoints of UniMate, a text-conditioned flow-matching model that generates motion for skeletons of any topology: animals, humanoids and rigged objects.
 
 ## Models
 
 | Model | Architecture | Training data | Joints | Steps | Params¹ | Status |
 |---|---|---|---|---|---|---|
-| [`unimate_uniml3d_f60_v2`](unimate_uniml3d_f60_v2) | graph attention, AdaLN text | UniML3D [`faaa817`](https://huggingface.co/datasets/Linzhan/UniML3D/tree/faaa81773b315247b03f183548e3898dbec2ce80) (2026-09-27) | 5–70 | 100k | 74.1M | **Recommended** |
-| [`unimate_uniml3d_f60_v2_full_cross_attn`](unimate_uniml3d_f60_v2_full_cross_attn) | full attention, cross-attention text | UniML3D `faaa817` (2026-09-27) | 5–70 | 100k | 66.2M | Variant |
-| [`unimate_uniml3d_f60_preview`](unimate_uniml3d_f60_preview) | graph attention, AdaLN text | UniML3D, pre-release build | 5–60 | 120k | 74.1M | Superseded |
-| [`unimate_mixamo_f60`](unimate_mixamo_f60) | graph attention, AdaLN text | UniML3D `faaa817` (2026-09-27), Mixamo only | 22 (one rig) | 120k | 47.8M | **Recommended for Mixamo humanoids** |
+| [`unimate_uniml3d_f60_v3_preview`](unimate_uniml3d_f60_v3_preview) | graph attention, AdaLN text | UniML3D [`d0f19d9`](https://huggingface.co/datasets/Linzhan/UniML3D/tree/d0f19d943038fb782a26c2a615bd9b3d4f3a8728) | 5–70 | 100k | 74.1M | **Recommended** |
+| [`unimate_uniml3d_f60_v2`](unimate_uniml3d_f60_v2) | graph attention, AdaLN text | UniML3D [`faaa817`](https://huggingface.co/datasets/Linzhan/UniML3D/tree/faaa81773b315247b03f183548e3898dbec2ce80) | 5–70 | 100k | 74.1M | Previous version |
+| [`unimate_uniml3d_f60_v2_cross_attn`](unimate_uniml3d_f60_v2_cross_attn) | full attention, cross-attention text | UniML3D `faaa817` | 5–70 | 100k | 66.2M | Variant |
+| [`unimate_mixamo_f60_v2`](unimate_mixamo_f60_v2) | graph attention, AdaLN text | UniML3D `faaa817`, Mixamo | 22 (one rig) | 120k | 47.8M | **Recommended for Mixamo** |
+| [`unimate_truebones_f60_v2`](unimate_truebones_f60_v2) | graph attention, AdaLN text | UniML3D `faaa817`, Truebones | 5–100 | 80k | 47.8M | **Recommended for Truebones** |
+| [`unimate_uniml3d_f60_v1_preview`](unimate_uniml3d_f60_v1_preview) | graph attention, AdaLN text | UniML3D pre-release build | 5–60 | 120k | 74.1M | Superseded |
 
-¹ Denoiser only. The frozen `google/flan-t5-base` text encoder is downloaded from the Hub on first use.
+¹ Denoiser only; the frozen `google/flan-t5-base` text encoder is downloaded on first use.
 
-Use each model's final checkpoint, `checkpoints/checkpoint_step_<steps>.pt`.
-
-**v2 versus preview.** Raising the joint limit from 60 to 70 keeps 98.1% of the training clips instead of 86.9%, and 70 of the 74 Truebones species. Datasets are balanced before object types (`sampler_dataset_alpha = 0.25`), so Truebones, Mixamo and Objaverse-XL make up about 13%, 24% and 63% of the training samples; the single Mixamo rig previously got under 1%.
-
-**Mixamo humanoids.** To test humanoid animation on the Mixamo rig, use `unimate_mixamo_f60`. It is trained on the 2,162 Mixamo clips alone, all on the one 22-joint rig, without topology augmentations. It animates that rig only; use a UniML3D model for any other skeleton.
+The Mixamo model animates the 22-joint Mixamo rig only. The Truebones model covers 73 of the 74 species (all but Dragon), including Bear, Centipede and Monkey, which exceed the UniML3D models' limit. Use each model's latest checkpoint.
 
 ## Model details
 
 | | |
 |---|---|
-| **Architecture** | Transformer denoiser, 10 layers (6 for `unimate_mixamo_f60`), width 512, 8 heads; flow matching with a linear path and velocity prediction |
-| **Inputs** | English motion description; the target skeleton's T-pose, joint hierarchy and joint names |
-| **Output** | 60 frames at 30 fps, 12 features per joint: 3-D position, 6-D rotation relative to the T-pose, 3-D velocity. The root joint encodes height, facing and planar velocity instead |
-| **Text encoder** | `google/flan-t5-base`, frozen |
-| **Guidance** | Classifier-free guidance on the caption: 10% caption dropout in training, default scale 3.0 |
+| **Architecture** | Transformer denoiser, width 512, 8 heads, 10 layers (6 for the Mixamo and Truebones models); flow matching, linear path, velocity prediction |
+| **Inputs** | English motion description; the skeleton's T-pose, hierarchy and cleaned joint labels |
+| **Output** | 60 frames at 30 fps; per joint: position (3), rotation relative to the T-pose (6), velocity (3) |
+| **Text** | Frozen `google/flan-t5-base`; classifier-free guidance (10% caption dropout, default scale 3.0) |
 
 ## Quick start
 
-Run every command from the root of the [code repository](https://github.com/Friedrich-M/UniMate), with its `unimate` environment installed.
+From the root of the [code repository](https://github.com/Friedrich-M/UniMate), in its `unimate` environment:
+
+```bash
+# 1. download a model (for another model, change the folder and the step)
+hf download Linzhan/UniMate --local-dir outputs \
+    --include "unimate_uniml3d_f60_v3_preview/*.json" --include "unimate_uniml3d_f60_v3_preview/*.npy" \
+    --include "unimate_uniml3d_f60_v3_preview/checkpoints/checkpoint_step_100000.pt"
+
+# 2. prepare a rigged GLB / glTF / FBX, animated or not: label its joints, check REVIEW.md, then build
+python -m data_process.rig_preprocess run --input robot.glb --output_dir outputs/rig/robot
+python -m data_process.rig_preprocess run --input robot.glb --output_dir outputs/rig/robot \
+    --annotation outputs/rig/robot/annotation.json
+
+# 3. generate: every prompt on every asset (dataset skeletons by name, e.g. truebones:Horse)
+python -m unimate.inference.sample --exp_dir outputs/unimate_uniml3d_f60_v3_preview \
+    --asset outputs/rig/robot truebones:Horse --prompt "An object walks forward." --num_repetitions 3
+
+# 4. drive each asset's mesh with its motions (one GLB / FBX per motion)
+bash scripts/run_animate_motion.sh outputs/unimate_uniml3d_f60_v3_preview/samples
+```
 
-**1. Build the dataset features.** The sampler reads each target skeleton from `dataset/features/<dataset>/`. The Hub dataset ships the stage 1-3 export but not these stage-4 features, so build them from the revision the model was trained on:
+A dataset skeleton needs only its source's conditioning, plus canonical meshes to drive:
 
 ```bash
-hf download Linzhan/UniML3D --repo-type dataset \
-    --revision faaa81773b315247b03f183548e3898dbec2ce80 --local-dir dataset
-bash data_process/scripts/run_extract_features.sh truebones   # likewise mixamo, objaverse
+hf download Linzhan/UniML3D --repo-type dataset --local-dir dataset --include "features/*/cond.npy" --include "canonical_assets/mixamo/*"
 ```
 
-`unimate_mixamo_f60` needs only the `mixamo` features.
+Test-case files, in-betweening, joint editing and motion expansion are documented in the code README.
 
-The Truebones motions come from a commercial pack and are not on the Hub; the dataset card explains how to rebuild them.
+## Training
 
-**2. Download a model.** This fetches the configuration, normalization statistics and final checkpoint of the recommended model:
+AdamW (lr 1e-4, betas 0.9 / 0.99, weight decay 1e-5, clipping 1.0), 3% linear warmup then cosine decay to 5%, EMA 0.9999 (used at inference). Loss: masked L2 flow matching + 0.5 × geodesic rotation + 0.1 × velocity smoothness. 60-frame windows at 30 fps, batch 16 per GPU, with joint addition / removal, pooling and perturbation augmentations (Mixamo model: batch 32, no augmentation).
 
-```bash
-hf download Linzhan/UniMate \
-    --include "unimate_uniml3d_f60_v2/*.json" "unimate_uniml3d_f60_v2/*.npy" \
-              "unimate_uniml3d_f60_v2/checkpoints/checkpoint_step_100000.pt" \
-    --local-dir outputs
-```
+| Model | GPUs | Time |
+|---|---|---|
+| `unimate_uniml3d_f60_v3_preview` | 8× H100 | ≈ 23 h |
+| `unimate_uniml3d_f60_v2` | 8× H100 | ≈ 22 h |
+| `unimate_uniml3d_f60_v2_cross_attn` | 8× H100 | ≈ 36 h |
+| `unimate_mixamo_f60_v2` | 4× H100 | ≈ 12 h |
+| `unimate_truebones_f60_v2` | 4× H100 | ≈ 15 h |
+| `unimate_uniml3d_f60_v1_preview` | 6× H100 | ≈ 23 h |
 
-To test Mixamo humanoid animation, fetch `unimate_mixamo_f60` instead:
+Each model's `config.json` is its full training configuration. The code repository's `configs/uniml3d_60frames_graph_adaln.json` follows the v3_preview recipe, and `configs/{mixamo,truebones}_60frames_graph_adaln.json` those of the Mixamo and Truebones models:
 
 ```bash
-hf download Linzhan/UniMate \
-    --include "unimate_mixamo_f60/*.json" "unimate_mixamo_f60/*.npy" \
-              "unimate_mixamo_f60/checkpoints/checkpoint_step_120000.pt" \
-    --local-dir outputs
+accelerate launch --num_processes 8 -m unimate.training.train --config configs/uniml3d_60frames_graph_adaln.json
 ```
 
-**3. Generate motion from text.** Write the prompts to `test_cases.json` with keys `<object_type>-<case_id>`. `object_type` is a skeleton in `dataset/features/`: a Truebones species, `mixamo`, or an Objaverse-XL object ID. `case_id` is a free-form tag that names the output files. With `unimate_mixamo_f60`, every key is `mixamo-<case_id>` and `--exp_dir` is `outputs/unimate_mixamo_f60`.
+## Repository layout
 
-```json
-{
-  "Horse-0": "An object rears up on its hind legs.",
-  "mixamo-0": "An object jumps in place with both arms raised.",
-  "144367de23534c28ad2e83fc8abbd9ea-0": "An object flaps its wings."
-}
 ```
-
-```bash
-python -m unimate.inference.sample \
-    --exp_dir outputs/unimate_uniml3d_f60_v2 \
-    --test_cases_json test_cases.json \
-    --num_repetitions 3
+config.json                    index of the models
+<model>/
+  config.json                  model and data configuration, read by inference
+  dataset_stats.npy            normalization statistics
+  checkpoints/                 checkpoint_step_<N>.pt every 10k steps (model, EMA, optimizer, scheduler)
+  logs/                        TensorBoard event files
+  samples/step_<NNNNNN>/       skeleton renders of motions sampled during training
 ```
 
-The sampler loads the EMA weights of the latest checkpoint in `checkpoints/`. It writes the motions (`.npy`, shape `(60, J, 12)` for a skeleton with `J` joints) and skeleton renders (`.mp4`) to `outputs/unimate_uniml3d_f60_v2/samples/`. In-betweening, joint-level editing, motion expansion and driving a rigged mesh are documented in the code README.
+`unimate_uniml3d_f60_v3_preview` holds its latest three checkpoints, without logs or samples.
 
-## Training
-
-| | |
-|---|---|
-| **Optimizer** | AdamW, learning rate 1e-4, betas (0.9, 0.99), weight decay 1e-5, gradient clipping at 1.0 |
-| **Schedule** | Linear warmup over the first 3% of steps, then cosine decay to 5% of the peak rate |
-| **Loss** | Masked L2 flow-matching loss + 0.5 × geodesic rotation loss + 0.1 × velocity smoothness loss |
-| **EMA** | Decay 0.9999; used at inference |
-| **Batch size** | 16 per GPU (32 for `unimate_mixamo_f60`) |
-| **Data** | 60-frame windows at 30 fps; joint addition, joint removal, pooling and perturbation augmentations (off for `unimate_mixamo_f60`) |
+## Versions
 
-| Model | Config (code repository) | GPUs | Training time |
-|---|---|---|---|
-| `unimate_uniml3d_f60_v2` | `configs/uniml3d_60frames_graph_adaln_v2.json` | 8× H100 | about 22 hours |
-| `unimate_uniml3d_f60_v2_full_cross_attn` | `configs/uniml3d_60frames_full_cross_attn_v2.json` | 8× H100 | about 36 hours² |
-| `unimate_uniml3d_f60_preview` | `configs/uniml3d_60frames_graph_adaln.json` | 6× H100 | about 23 hours |
-| `unimate_mixamo_f60` | `configs/mixamo_60frames_graph_adaln.json` | 4× H100 | about 12 hours |
+- **v3_preview** (2026-10-05), over v2: newer data (UniML3D `d0f19d9`: re-reviewed annotations; 480 Objaverse-XL rigs authored lying down or upside down stood upright instead of filtered); three captions per clip (generic 35%, detailed 25%, normal 40%); fixed per-dataset sampling weights; one normalization pool for all datasets; logit-normal flow time.
+- **v2** (2026-09-28), over v1_preview: UniML3D `faaa817`; joint limit 70 instead of 60 (98.1% of the clips instead of 86.9%); datasets balanced before object types (Mixamo from under 1% to about 24% of the samples); 100k steps. Includes the cross-attention variant and the Mixamo and Truebones models.
+- **v1_preview** (2026-09-27): first release, on a pre-release build of UniML3D.
 
-² Resumed once from the step-60k checkpoint.
+## Limitations
 
-Resume from a released checkpoint, or train from scratch:
+- **Contact.** No unified contact model: contact-rich motions can slide, drift, hover or penetrate the ground; foot locking or IK post-processing helps where contacts are well defined.
+- **Rare topologies and motions.** The data is long-tailed toward humanoids and common locomotion; rare skeletons and unusual motions can come out static, jittery or off-prompt.
+- **Format.** Fixed 60-frame (2 s) samples (longer motion via expansion); a skeleton must fit the model's joint limit; prompts should describe the motion only, as every training caption starts with "An object".
 
-```bash
-bash scripts/run_train.sh configs/uniml3d_60frames_graph_adaln_v2.json -- \
-    --resume outputs/unimate_uniml3d_f60_v2/checkpoints/checkpoint_step_100000.pt
+See the [paper](https://arxiv.org/abs/2609.05415) (Section 6, Appendix F) for details.
 
-accelerate launch --num_processes 8 -m unimate.training.train \
-    --config configs/uniml3d_60frames_graph_adaln_v2.json
-```
+## License
 
-## Repository layout
+Checkpoints: [CC BY-NC 4.0](LICENSE). Code: MIT. This covers only our rights in the weights; the training data keep their source terms:
 
-```
-config.json                        index of the released models
-LICENSE
-<model>/
-  config.json                      resolved training configuration; read by inference
-  dataset_stats.npy                feature normalization statistics
-  checkpoints/
-    checkpoint_step_<N>.pt         model and EMA weights, optimizer and scheduler state; every 10k steps
-  logs/                            TensorBoard curves, one event file per training job
-  samples/
-    step_<NNNNNN>/                 motions generated during training (step_000000: before training)
-      <object_type>-<i>_fk.mp4     joints from forward kinematics of the predicted rotations
-      <object_type>-<i>_ric.mp4    predicted joint positions
-```
+| Model | Training data |
+|---|---|
+| `unimate_uniml3d_f60_v3_preview`, `_v2`, `_v2_cross_attn`, `_v1_preview` | Truebones ZOO, Mixamo, Objaverse-XL |
+| `unimate_mixamo_f60_v2` | Mixamo |
+| `unimate_truebones_f60_v2` | Truebones ZOO |
 
-## Limitations
+- **Truebones ZOO**: a commercial asset pack by [Truebones](https://truebones.com). Third parties allege that some of its animal assets come from commercial games (among them Resident Evil, Skyrim, Bless Online and Conan Exiles). We have not verified these claims; until they are resolved, treat the provenance of the Truebones-trained models as unclear.
+- **Mixamo**: Adobe's Mixamo terms of use.
+- **Objaverse-XL**: the license of each source object; some are non-commercial.
 
-- **Clip length:** a sample is a fixed 60-frame window (2 seconds). Longer motion requires chaining samples with the expansion application.
-- **Joint count:** the v2 models are trained on skeletons with at most 70 joints, which excludes four Truebones species (Bear, Centipede, Monkey, Dragon); the preview model's limit is 60. Skeletons with more than 71 joints (61 for the preview model) cannot be sampled. `unimate_mixamo_f60` has seen only the 22-joint Mixamo rig and cannot sample a larger skeleton.
-- **New skeletons:** there is no skeleton-only input. A new rig must be converted into a feature directory with `data_process/scripts/run_preprocess_char.sh`, and needs at least one animation clip.
-- **Text:** every training caption has the subject "An object"; the skeleton determines the character. Prompts should describe the motion only.
+We make no warranty that the checkpoints, or motions generated with them, are free of third-party rights. Rights holders with a concern can open a discussion on this repository or contact the authors.
 
 ## Citation
 
@@ -169,20 +153,3 @@ LICENSE
   year    = {2026}
 }
 ```
-
-## License
-
-The checkpoints are released under the [Creative Commons Attribution-NonCommercial 4.0 International License](LICENSE) (CC BY-NC 4.0). The UniMate [code](https://github.com/Friedrich-M/UniMate) remains under the MIT License.
-
-This license covers only our own rights in the weights. It grants no rights in the training data, which remain under their source terms:
-
-| Model | Training data | Terms |
-|---|---|---|
-| `unimate_uniml3d_f60_v2`, `unimate_uniml3d_f60_v2_full_cross_attn`, `unimate_uniml3d_f60_preview` | Truebones ZOO, Mixamo, Objaverse-XL | all three below |
-| `unimate_mixamo_f60` | Mixamo only | Adobe Mixamo |
-
-- **Truebones ZOO:** a commercial asset pack sold by [Truebones](https://truebones.com). Third parties have alleged that some of its animal assets come from commercial games (among them Resident Evil, Skyrim, Bless Online and Conan Exiles). We have not verified these claims. Until they are resolved, treat the provenance of the Truebones-trained models as unclear.
-- **Mixamo:** Adobe's Mixamo terms of use.
-- **Objaverse-XL:** the license attached to each source object. Some of these licenses are non-commercial.
-
-The authors make no warranty that the checkpoints, or the motions generated with them, are free of third-party rights. If you hold rights in any of the training data and have a concern, please open a discussion on this repository or contact the authors.

config.json

@@ -3,9 +3,9 @@
   "code": "https://github.com/Friedrich-M/UniMate",
   "paper": "https://arxiv.org/abs/2609.05415",
   "models": {
-    "unimate_uniml3d_f60_v2": {
+    "unimate_uniml3d_f60_v3_preview": {
       "dataset": "Linzhan/UniML3D",
-      "train_config": "configs/uniml3d_60frames_graph_adaln_v2.json",
+      "train_config": "configs/uniml3d_60frames_graph_adaln.json",
       "attention": "graph",
       "text_cond": "adaln",
       "num_layers": 10,
@@ -15,11 +15,25 @@
       "steps": 100000,
       "recommended_checkpoint": "checkpoints/checkpoint_step_100000.pt",
       "status": "recommended",
+      "dataset_revision": "d0f19d943038fb782a26c2a615bd9b3d4f3a8728"
+    },
+    "unimate_uniml3d_f60_v2": {
+      "dataset": "Linzhan/UniML3D",
+      "train_config": "unimate_uniml3d_f60_v2/config.json",
+      "attention": "graph",
+      "text_cond": "adaln",
+      "num_layers": 10,
+      "latent_dim": 512,
+      "num_frames": 60,
+      "fps": 30,
+      "steps": 100000,
+      "recommended_checkpoint": "checkpoints/checkpoint_step_100000.pt",
+      "status": "previous",
       "dataset_revision": "faaa81773b315247b03f183548e3898dbec2ce80"
     },
-    "unimate_uniml3d_f60_v2_full_cross_attn": {
+    "unimate_uniml3d_f60_v2_cross_attn": {
       "dataset": "Linzhan/UniML3D",
-      "train_config": "configs/uniml3d_60frames_full_cross_attn_v2.json",
+      "train_config": "unimate_uniml3d_f60_v2_cross_attn/config.json",
       "attention": "full",
       "text_cond": "cross_attn",
       "num_layers": 10,
@@ -31,32 +45,46 @@
       "status": "variant",
       "dataset_revision": "faaa81773b315247b03f183548e3898dbec2ce80"
     },
-    "unimate_uniml3d_f60_preview": {
+    "unimate_mixamo_f60_v2": {
       "dataset": "Linzhan/UniML3D",
-      "train_config": "configs/uniml3d_60frames_graph_adaln.json",
+      "train_config": "configs/mixamo_60frames_graph_adaln.json",
       "attention": "graph",
       "text_cond": "adaln",
-      "num_layers": 10,
+      "num_layers": 6,
       "latent_dim": 512,
       "num_frames": 60,
       "fps": 30,
       "steps": 120000,
       "recommended_checkpoint": "checkpoints/checkpoint_step_120000.pt",
-      "status": "superseded"
+      "status": "recommended_for_mixamo",
+      "dataset_revision": "faaa81773b315247b03f183548e3898dbec2ce80"
     },
-    "unimate_mixamo_f60": {
+    "unimate_truebones_f60_v2": {
       "dataset": "Linzhan/UniML3D",
-      "train_config": "configs/mixamo_60frames_graph_adaln.json",
+      "train_config": "configs/truebones_60frames_graph_adaln.json",
       "attention": "graph",
       "text_cond": "adaln",
       "num_layers": 6,
       "latent_dim": 512,
       "num_frames": 60,
       "fps": 30,
+      "steps": 80000,
+      "recommended_checkpoint": "checkpoints/checkpoint_step_80000.pt",
+      "status": "recommended_for_truebones",
+      "dataset_revision": "faaa81773b315247b03f183548e3898dbec2ce80"
+    },
+    "unimate_uniml3d_f60_v1_preview": {
+      "dataset": "Linzhan/UniML3D",
+      "train_config": "unimate_uniml3d_f60_v1_preview/config.json",
+      "attention": "graph",
+      "text_cond": "adaln",
+      "num_layers": 10,
+      "latent_dim": 512,
+      "num_frames": 60,
+      "fps": 30,
       "steps": 120000,
       "recommended_checkpoint": "checkpoints/checkpoint_step_120000.pt",
-      "status": "recommended_for_mixamo",
-      "dataset_revision": "faaa81773b315247b03f183548e3898dbec2ce80"
+      "status": "superseded"
     }
   }
 }