Instructions to use wsagi/DiffusionPolicy-PickOrange with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use wsagi/DiffusionPolicy-PickOrange with LeRobot:
- Notebooks
- Google Colab
- Kaggle
Hot-swap noise scheduler: DDPM 100-step → DDIM 32-step (no retraining)
Browse files- config.json +2 -2
config.json
CHANGED
|
@@ -71,7 +71,7 @@
|
|
| 71 |
"n_groups": 8,
|
| 72 |
"diffusion_step_embed_dim": 128,
|
| 73 |
"use_film_scale_modulation": true,
|
| 74 |
-
"noise_scheduler_type": "
|
| 75 |
"num_train_timesteps": 100,
|
| 76 |
"beta_schedule": "squaredcos_cap_v2",
|
| 77 |
"beta_start": 0.0001,
|
|
@@ -79,7 +79,7 @@
|
|
| 79 |
"prediction_type": "epsilon",
|
| 80 |
"clip_sample": true,
|
| 81 |
"clip_sample_range": 1.0,
|
| 82 |
-
"num_inference_steps":
|
| 83 |
"compile_model": false,
|
| 84 |
"compile_mode": "reduce-overhead",
|
| 85 |
"do_mask_loss_for_padding": false,
|
|
|
|
| 71 |
"n_groups": 8,
|
| 72 |
"diffusion_step_embed_dim": 128,
|
| 73 |
"use_film_scale_modulation": true,
|
| 74 |
+
"noise_scheduler_type": "DDIM",
|
| 75 |
"num_train_timesteps": 100,
|
| 76 |
"beta_schedule": "squaredcos_cap_v2",
|
| 77 |
"beta_start": 0.0001,
|
|
|
|
| 79 |
"prediction_type": "epsilon",
|
| 80 |
"clip_sample": true,
|
| 81 |
"clip_sample_range": 1.0,
|
| 82 |
+
"num_inference_steps": 32,
|
| 83 |
"compile_model": false,
|
| 84 |
"compile_mode": "reduce-overhead",
|
| 85 |
"do_mask_loss_for_padding": false,
|