Release TokenSwap-Bench
Browse files- README.md +2 -0
- assets/tokenswap_example.png +3 -0
README.md
CHANGED
|
@@ -41,6 +41,8 @@ configs:
|
|
| 41 |
|
| 42 |
</div>
|
| 43 |
|
|
|
|
|
|
|
| 44 |
<p align="center"><img src="assets/fig1_overview.png" width="80%"></p>
|
| 45 |
|
| 46 |
Across **42 MLLMs**, we observe a pervasive modality gap, with performance decreasing by **4.2% to 47.4%** when moving from text-only to image-interleaved inputs.
|
|
|
|
| 41 |
|
| 42 |
</div>
|
| 43 |
|
| 44 |
+
<p align="center"><img src="assets/tokenswap_example.png" width="90%"></p>
|
| 45 |
+
|
| 46 |
<p align="center"><img src="assets/fig1_overview.png" width="80%"></p>
|
| 47 |
|
| 48 |
Across **42 MLLMs**, we observe a pervasive modality gap, with performance decreasing by **4.2% to 47.4%** when moving from text-only to image-interleaved inputs.
|
assets/tokenswap_example.png
ADDED
|
Git LFS Details
|