Update README.md
Browse files
README.md
CHANGED
|
@@ -27,6 +27,8 @@ It is _**the largest open-source vision/vision-language foundation model (14B)**
|
|
| 27 |
- Params (M): 5903
|
| 28 |
- Image size: 224 x 224
|
| 29 |
- **Pretrain Dataset:** LAION-en, LAION-COCO, COYO, CC12M, CC3M, SBU, Wukong, LAION-multi
|
|
|
|
|
|
|
| 30 |
|
| 31 |
## Linear Probing Performance
|
| 32 |
|
|
|
|
| 27 |
- Params (M): 5903
|
| 28 |
- Image size: 224 x 224
|
| 29 |
- **Pretrain Dataset:** LAION-en, LAION-COCO, COYO, CC12M, CC3M, SBU, Wukong, LAION-multi
|
| 30 |
+
- **Note:** This model has 48 blocks, and we found that using the output after the fourth-to-last block worked best for VLLM. Therefore, **please set mm_vision_select_layer=-4 when using this model to build VLLM.**
|
| 31 |
+
|
| 32 |
|
| 33 |
## Linear Probing Performance
|
| 34 |
|