Release unimate_uniml3d_f60_v3 trained to 150k steps with checkpoints 100k-150k; update the model card and index

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

.gitignore

@@ -6,7 +6,5 @@
 # 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
+# unimate_uniml3d_f60_v3: training logs stay local
+unimate_uniml3d_f60_v3/logs/

README.md

@@ -29,11 +29,11 @@ Pretrained checkpoints of UniMate, a text-conditioned flow-matching model that g
 
 | Model | Architecture | Training data | Joints | Steps | Params¹ | Status |
 |---|---|---|---|---|---|---|
-| [`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_v3`](unimate_uniml3d_f60_v3) | graph attention, AdaLN text | UniML3D [`d0f19d9`](https://huggingface.co/datasets/Linzhan/UniML3D/tree/d0f19d943038fb782a26c2a615bd9b3d4f3a8728) | 5–70 | 150k | 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_truebones_f60_v2`](unimate_truebones_f60_v2) | graph attention, AdaLN text | UniML3D `faaa817`, Truebones | 5–85 | 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| graph attention, AdaLN text | UniML3D pre-release build | 5–60 | 120k | 74.1M | Superseded |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 on first use.
@@ -54,31 +54,21 @@ The Mixamo model animates the 22-joint Mixamo rig only. The Truebones model cove
 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. download a model (for another model or checkpoint, change the folder and the step)
+hf download Linzhan/UniMate --repo-type model --local-dir outputs \
+    --include "unimate_uniml3d_f60_v3/*.json" --include "unimate_uniml3d_f60_v3/*.npy" \
+    --include "unimate_uniml3d_f60_v3/checkpoints/checkpoint_step_150000.pt"
+
+# 2. generate for an example asset shipped with the code (assets/examples/ also has an eagle and a shark)
+python -m unimate.inference.sample --exp_dir outputs/unimate_uniml3d_f60_v3 --asset assets/examples/unitree_go2 \
+    --prompt "An object trots forward." "An object rears up on its hind legs." --num_repetitions 3 \
+    --output_dir outputs/samples/custom
+
+# 3. drive each asset's mesh with its motions (one GLB / FBX per motion)
+bash scripts/run_animate_motion.sh outputs/samples/custom
 ```
 
-A dataset skeleton needs only its source's conditioning, plus canonical meshes to drive:
-
-```bash
-hf download Linzhan/UniML3D --repo-type dataset --local-dir dataset --include "features/*/cond.npy" --include "canonical_assets/mixamo/*"
-```
-
-Test-case files, in-betweening, joint editing and motion expansion are documented in the code README.
+For your own rigged GLB / glTF / FBX, `python -m data_process.rig_preprocess run --input <asset> --output_dir outputs/rig/<name>` labels its joints, stops so you can check them, and builds an asset directory that `--asset` takes. Skeletons of the [UniML3D](https://huggingface.co/datasets/Linzhan/UniML3D) dataset are named `<dataset>:<object_type>` (e.g. `mixamo`, `objaverse:<uid>`) once it is downloaded into `dataset/`. The code README covers both, along with prompting, in-betweening, joint editing and motion expansion.
 
 ## Training
 
@@ -86,14 +76,14 @@ AdamW (lr 1e-4, betas 0.9 / 0.99, weight decay 1e-5, clipping 1.0), 3% linear wa
 
 | Model | GPUs | Time |
 |---|---|---|
-| `unimate_uniml3d_f60_v3_preview` | 8× H100 | ≈ 23 h |
+| `unimate_uniml3d_f60_v3` | 8× H100 | ≈ 35 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 |
 
-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:
+Each model's `config.json` holds the model and data configuration that inference reads. The code repository's `configs/uniml3d_60frames_graph_adaln.json` follows the v3 recipe (at 120k steps; v3 trained 150k), and `configs/{mixamo,truebones}_60frames_graph_adaln.json` those of the Mixamo and Truebones models:
 
 ```bash
 accelerate launch --num_processes 8 -m unimate.training.train --config configs/uniml3d_60frames_graph_adaln.json
@@ -111,11 +101,11 @@ config.json                    index of the models
   samples/step_<NNNNNN>/       skeleton renders of motions sampled during training
 ```
 
-`unimate_uniml3d_f60_v3_preview` holds its latest three checkpoints, without logs or samples.
+`unimate_uniml3d_f60_v3` holds the checkpoints from 100k to 150k steps (150k recommended), without logs or samples.
 
 ## Versions
 
-- **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.
+- **v3** (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; 150k steps.
 - **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.
 
@@ -133,7 +123,7 @@ Checkpoints: [CC BY-NC 4.0](LICENSE). Code: MIT. This covers only our rights in
 
 | Model | Training data |
 |---|---|
-| `unimate_uniml3d_f60_v3_preview`, `_v2`, `_v2_cross_attn`, `_v1_preview` | Truebones ZOO, Mixamo, Objaverse-XL |
+| `unimate_uniml3d_f60_v3`, `_v2`, `_v2_cross_attn`, `_v1_preview` | Truebones ZOO, Mixamo, Objaverse-XL |
 | `unimate_mixamo_f60_v2` | Mixamo |
 | `unimate_truebones_f60_v2` | Truebones ZOO |
 

config.json

@@ -3,7 +3,7 @@
   "code": "https://github.com/Friedrich-M/UniMate",
   "paper": "https://arxiv.org/abs/2609.05415",
   "models": {
-    "unimate_uniml3d_f60_v3_preview": {
+    "unimate_uniml3d_f60_v3": {
       "dataset": "Linzhan/UniML3D",
       "train_config": "configs/uniml3d_60frames_graph_adaln.json",
       "attention": "graph",
@@ -12,8 +12,8 @@
       "latent_dim": 512,
       "num_frames": 60,
       "fps": 30,
-      "steps": 100000,
-      "recommended_checkpoint": "checkpoints/checkpoint_step_100000.pt",
+      "steps": 150000,
+      "recommended_checkpoint": "checkpoints/checkpoint_step_150000.pt",
       "status": "recommended",
       "dataset_revision": "d0f19d943038fb782a26c2a615bd9b3d4f3a8728"
     },

unimate_uniml3d_f60_v3_preview/checkpoints/checkpoint_step_80000.pt → unimate_uniml3d_f60_v3/checkpoints/checkpoint_step_110000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e6006df925fdeed2d305740424a0ec8fe78a43769b449f176336e5d88eb8e822
-size 1185834029
+oid sha256:f2d78b1e398d70eff38192f6fb4679b4c2dd956ccd1589076a0bd155a1c90d4e
+size 1185836040

unimate_uniml3d_f60_v3_preview/checkpoints/checkpoint_step_90000.pt → unimate_uniml3d_f60_v3/checkpoints/checkpoint_step_120000.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c7ed43ab5c370c69b2ae3a1bf882fc326af412df3deca8ad614df2dba6e67e22
-size 1185834029
+oid sha256:93acd9af9f11e6afe1c6dcadaa86d065887f431865e9db3fd56d918015222662
+size 1185836040

unimate_uniml3d_f60_v3/checkpoints/checkpoint_step_130000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c61ad69f9a3cb7b4213de3b7fab7fe297c7853a7a5f12b03a7c5b196c3c9c44
+size 1185836040

unimate_uniml3d_f60_v3/checkpoints/checkpoint_step_140000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6969ae36f6329e84d3ed8ed4ecb7758e6fc212ceeba28a816c777c560fde4d53
+size 1185836040

unimate_uniml3d_f60_v3/checkpoints/checkpoint_step_150000.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f991ddd87262c251f4681d18297a6163ab4f139daa28c86d5ef0a66771df7bc6
+size 1185836040