Card: pruning diagram and speed/error trade-off figure

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/tradeoff.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -36,8 +36,16 @@ Test sets; greedy CTC decoding, no language model. CER for Khmer, WER for Englis
 - **English:** it keeps much less of its teacher's accuracy, with about 1.5–2× the teacher's WER.
 - **In-domain test sets:** FLEURS-km and FLEURS-en train are in the training mix, so their test sets are in-domain.
 
+<p align="center"><img src="https://huggingface.co/Darayut/sota-base/resolve/main/assets/tradeoff.png" width="100%" alt="Error rate against speed-up for each fine-tuned teacher and its 80%-sparse model"></p>
+
+*Speed against error.* Each arrow runs from a fine-tuned teacher to its pruned model; speed is relative to the fine-tuned 1B (real-time factor).
+
 ## Architecture
 
+<p align="center"><img src="https://huggingface.co/Darayut/sota-base/resolve/main/assets/pruning.png" width="100%" alt="The encoder before and after pruning: CNN kept, attention heads and FFN units removed"></p>
+
+*What is pruned*, shown for sota-large: whole attention heads (MHA) and FFN units are removed from every transformer layer, and the CNN front end is kept. Each cell is a quarter of a layer's heads or units.
+
 - **Base:** a wav2vec2-style CTC encoder with Omnilingual's 9,812-entry character vocabulary, which covers Khmer and Latin script.
 - **Pruned parts:** per-layer attention heads and FFN units, so every layer has its own width. `config.json` lists the exact `num_heads` and `ffn_dims`; the encoder has 63M parameters.
 - **Unchanged parts:** the CNN feature extractor and the output layer.