Update README.md

README.md

@@ -80,7 +80,23 @@ SmolVLM2 is built upon [SigLIP](https://huggingface.co/google/siglip-base-patch1
 
 We release the SmolVLM 2checkpoints under the Apache 2.0 license.
 
-## Training Details
+## Training Data
 
-### Training Data
+SmolVLM2 used 3.3M samples for training coming from ten datasets: LlaVa Onevision, M4-Instruct, Mammoth, LlaVa Video 178K, FineVideo, VideoStar, VRipt, Vista-400K, MovieChat and ShareGPT4Video.
+
+### General split
+
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolvlm2_data_split.png" width="800" height="auto" alt="Image description">
+
+### Text mixture
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolvlm2_data_text.png" width="800" height="auto" alt="Image description">
+
+### Image mixture
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolvlm2_data_image.png" width="800" height="auto" alt="Image description">
+
+### Multi-image mixture
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolvlm2_data_multiimage.png" width="800" height="auto" alt="Image description">
+
+### Video mixture
+<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/smolvlm2_data_video.png" width="800" height="auto" alt="Image description">