Ferry1231 commited on
Commit ·
80410e0
1
Parent(s): ce0fdd0
update model card
Browse files
README.md
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
-
## license: apache-2.0
|
| 4 |
library_name: diffusers
|
| 5 |
tags:
|
| 6 |
- text-to-image
|
|
@@ -15,6 +14,7 @@ tags:
|
|
| 15 |
base_model:
|
| 16 |
- black-forest-labs/FLUX.1-dev
|
| 17 |
- Qwen/Qwen-Image-Edit
|
|
|
|
| 18 |
|
| 19 |
# ARR-RPO
|
| 20 |
|
|
@@ -24,8 +24,8 @@ base_model:
|
|
| 24 |
|
| 25 |
ARR-RPO provides two LoRA adapters trained with **Auto-Rubric as Reward (ARR)** and **Rubric Policy Optimization (RPO)** for visual generation:
|
| 26 |
|
| 27 |
-
-
|
| 28 |
-
-
|
| 29 |
|
| 30 |
ARR-RPO uses a frozen VLM judge conditioned on explicit auto-generated rubrics. During RPO training, two candidate outputs are sampled for the same prompt or edit instruction, the ARR judge selects the preferred output, and the preferred/dispreferred candidates receive binary rewards. The goal is to improve prompt faithfulness, visual quality, compositional alignment, and edit fidelity without training a separate scalar reward model.
|
| 31 |
|
|
|
|
| 1 |
---
|
| 2 |
+
license: apache-2.0
|
|
|
|
| 3 |
library_name: diffusers
|
| 4 |
tags:
|
| 5 |
- text-to-image
|
|
|
|
| 14 |
base_model:
|
| 15 |
- black-forest-labs/FLUX.1-dev
|
| 16 |
- Qwen/Qwen-Image-Edit
|
| 17 |
+
---
|
| 18 |
|
| 19 |
# ARR-RPO
|
| 20 |
|
|
|
|
| 24 |
|
| 25 |
ARR-RPO provides two LoRA adapters trained with **Auto-Rubric as Reward (ARR)** and **Rubric Policy Optimization (RPO)** for visual generation:
|
| 26 |
|
| 27 |
+
- **`ARR-FLUX.1-dev/`**: a LoRA adapter for FLUX.1-dev text-to-image generation.
|
| 28 |
+
- **`ARR-Qwen-Image-Edit/`**: a LoRA adapter for Qwen-Image-Edit instruction-guided image editing.
|
| 29 |
|
| 30 |
ARR-RPO uses a frozen VLM judge conditioned on explicit auto-generated rubrics. During RPO training, two candidate outputs are sampled for the same prompt or edit instruction, the ARR judge selects the preferred output, and the preferred/dispreferred candidates receive binary rewards. The goal is to improve prompt faithfulness, visual quality, compositional alignment, and edit fidelity without training a separate scalar reward model.
|
| 31 |
|