Update README.md

README.md

@@ -1,2 +1,14 @@
 **POntAvignon** (Policy Ontology for Avignon festival) is small reasoning model based on Pleias-350m, specialized for the transformation of unstructured theatre program into structured semantic data.
 
+## Training
+
+POntAvignon is based on variant of Pleias-350m mid-trained on Wikidata with generalist abilities for the transformation of unstructured text into knowledge graphs. 
+
+The model has been further trained on samples of Avignon show programs with a reinforcement learning method (GRPO). We initially focused on constraining the model to only generate items about shows and use a limited set of properties. A judge model (Gemma 12B) contributed to enhance the overall accuracy of the results by assessing the preliminary drafts at each step.
+
+## Use
+PontAvignon is a specialized reasoning model using a special tokens to encode inputs, structured data outputs and the intermediary reasoning traces.
+
+Any excerpts of a show program has to be submitted using the tags '<|text_startl>' and '<|text_end|>'.
+
+