Update README.md
Browse files
README.md
CHANGED
|
@@ -3,6 +3,9 @@ license: apache-2.0
|
|
| 3 |
base_model:
|
| 4 |
- lightonai/LightOnOCR-2-1B
|
| 5 |
base_model_relation: adapter
|
|
|
|
|
|
|
|
|
|
| 6 |
---
|
| 7 |
|
| 8 |
# Image As Text For Flux.2 Klein
|
|
@@ -12,7 +15,7 @@ It is transformed into Qwen3 embeddings, which represent the underlying meaning
|
|
| 12 |
|
| 13 |
Compared to T5/Gemma/Qwen, the Mistral ViT model's output sequence length is considerably shorter. The same information is encoded more efficiently.
|
| 14 |
|
| 15 |
-
In other literature (2601.14750,
|
| 16 |
|
| 17 |
## Inference
|
| 18 |
|
|
@@ -50,4 +53,4 @@ pipeline = Flux2KleinPipeline.from_pretrained('black-forest-labs/FLUX.2-klein-ba
|
|
| 50 |
- spatial-caption
|
| 51 |
- spright-coco
|
| 52 |
- z-image-ethnicity-test
|
| 53 |
-
- benchmarks from the Qwen-Image Technical Report
|
|
|
|
| 3 |
base_model:
|
| 4 |
- lightonai/LightOnOCR-2-1B
|
| 5 |
base_model_relation: adapter
|
| 6 |
+
tags:
|
| 7 |
+
- flux2
|
| 8 |
+
- flux-2-klein
|
| 9 |
---
|
| 10 |
|
| 11 |
# Image As Text For Flux.2 Klein
|
|
|
|
| 15 |
|
| 16 |
Compared to T5/Gemma/Qwen, the Mistral ViT model's output sequence length is considerably shorter. The same information is encoded more efficiently.
|
| 17 |
|
| 18 |
+
In other literature (2601.14750, Figure 3) the [safetensors](https://huggingface.co/nightknocker/flux2-klein-4b-lighting-text-encoder/blob/main/model.safetensors) file is called Visual Projection Head.
|
| 19 |
|
| 20 |
## Inference
|
| 21 |
|
|
|
|
| 53 |
- spatial-caption
|
| 54 |
- spright-coco
|
| 55 |
- z-image-ethnicity-test
|
| 56 |
+
- benchmarks from the Qwen-Image Technical Report
|