Instructions to use Cloth-splatters/folding-dynamics-gps with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Cloth-splatters/folding-dynamics-gps with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Cloth-splatters/folding-dynamics-gps", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
Upload training config
Browse files- config.yml +65 -0
config.yml
ADDED
|
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
data_dir: /scratch/project_2015260/data/fold/fold_generated_meshes_seed_1397.h5
|
| 2 |
+
output_dir: /scratch/project_2015260/outputs/dynamics_gps/2026-05-04/16-54-43_6616925
|
| 3 |
+
logging_dir: logs/
|
| 4 |
+
gradient_accumulation_steps: 2
|
| 5 |
+
mixed_precision: bf16
|
| 6 |
+
report_to: wandb
|
| 7 |
+
seed: 259
|
| 8 |
+
use_ema: false
|
| 9 |
+
gradient_checkpointing: false
|
| 10 |
+
scale_lr: false
|
| 11 |
+
allow_tf32: true
|
| 12 |
+
adam_beta1: 0.9
|
| 13 |
+
adam_beta2: 0.999
|
| 14 |
+
adam_weight_decay: 0.01
|
| 15 |
+
adam_epsilon: 1.0e-08
|
| 16 |
+
max_grad_norm: 1.0
|
| 17 |
+
per_gpu_batch_size: 256
|
| 18 |
+
learning_rate: 0.0003
|
| 19 |
+
num_workers: 8
|
| 20 |
+
compile_model: false
|
| 21 |
+
lr_scheduler: cosine
|
| 22 |
+
lr_warmup_steps: 2000
|
| 23 |
+
max_train_steps: 900000
|
| 24 |
+
resume_from_checkpoint: null
|
| 25 |
+
pretrained_model_name_or_path: null
|
| 26 |
+
checkpointing_steps: 2000
|
| 27 |
+
sampling_steps: 1000
|
| 28 |
+
checkpoints_total_limit: 10
|
| 29 |
+
exp_name: GPSDynamics_d256_H8_F5_L8_lr0.0003_bs256_fold_04052026_1656
|
| 30 |
+
num_train_epochs: 999999
|
| 31 |
+
config: configs/train_dynamics_gps.yaml
|
| 32 |
+
do_classifier_free_guidance: false
|
| 33 |
+
val_inference_steps: 50
|
| 34 |
+
model_cfg:
|
| 35 |
+
type: GPSDynamicsModel
|
| 36 |
+
num_prev_frames: 3
|
| 37 |
+
num_next_frames: 5
|
| 38 |
+
action_dim: 3
|
| 39 |
+
hidden_dim: 256
|
| 40 |
+
num_layers: 8
|
| 41 |
+
num_heads: 8
|
| 42 |
+
edge_feature_dim: 4
|
| 43 |
+
max_num_nodes: 160
|
| 44 |
+
dropout: 0.1
|
| 45 |
+
num_embeds_ada_norm: 1000
|
| 46 |
+
diffusion_cfg:
|
| 47 |
+
type: DDPM_DynamicsGPS
|
| 48 |
+
num_train_timesteps: 1000
|
| 49 |
+
beta_schedule: scaled_linear
|
| 50 |
+
snr_gamma: 5.0
|
| 51 |
+
history_noise_std: 0.0001
|
| 52 |
+
wandb_cfg:
|
| 53 |
+
project_name: cloth_sim_mujoco
|
| 54 |
+
entity: null
|
| 55 |
+
tags:
|
| 56 |
+
- gps
|
| 57 |
+
- dynamics
|
| 58 |
+
dataset_cfg:
|
| 59 |
+
type: ClothDynamicsVariableDataset
|
| 60 |
+
data_dir: /scratch/project_2015260/data/fold/fold_generated_meshes_seed_1397.h5
|
| 61 |
+
num_prev_frames: 3
|
| 62 |
+
num_next_frames: 5
|
| 63 |
+
max_num_nodes: 160
|
| 64 |
+
max_num_edges: 832
|
| 65 |
+
val_trajectory_ratio: 0.1
|