Model card: plain weight description
Browse files
README.md
CHANGED
|
@@ -16,7 +16,7 @@ Disen-Forcing is an autoregressive text-to-video model distilled from Wan2.1-T2V
|
|
| 16 |
|
| 17 |
## Model
|
| 18 |
|
| 19 |
-
- `disen_forcing.safetensors`:
|
| 20 |
- Student: Wan2.1-T2V-1.3B, initialized from the Causal Forcing chunk-wise ODE checkpoint; teacher: Wan2.1-T2V-14B.
|
| 21 |
- Training: DMD distillation with learnable context routing, EMA-based motion regularization and the KV caching gradient (see the [code repository](https://github.com/Dijkstra14/Disen-Forcing)).
|
| 22 |
|
|
|
|
| 16 |
|
| 17 |
## Model
|
| 18 |
|
| 19 |
+
- `disen_forcing.safetensors`: model weights (bf16).
|
| 20 |
- Student: Wan2.1-T2V-1.3B, initialized from the Causal Forcing chunk-wise ODE checkpoint; teacher: Wan2.1-T2V-14B.
|
| 21 |
- Training: DMD distillation with learnable context routing, EMA-based motion regularization and the KV caching gradient (see the [code repository](https://github.com/Dijkstra14/Disen-Forcing)).
|
| 22 |
|