Update README.md

README.md

@@ -3,10 +3,24 @@ library_name: transformers
 tags: []
 ---
 
-# Model Card for Model ID
-
-<!-- Provide a quick summary of what the model is/does. -->
-
+# Tutor Moves - Math Remediation: Provide a Problem-Specific Solution Strategy
+
+The tutor provides a strategy or next step for solving the problem. 
+
+## Message Structure
+
+Tutor CoPilot models are trained on a message structure of 10 context messages followed by one target message with the special tokens `[PRETEXT]` and `[TEXT]` used to demarcate the context and target messages. Messages are formatted as `{speaker}: {message}`, where the speaker is one of `tutor` or `student`, and the message is a lowercased, anonymized version of the message. Names are anonymized with [Edu-ConvoKit](https://edu-convokit.readthedocs.io/en/latest/preprocessing.html#edu_convokit.preprocessors.TextPreprocessor.anonymize_known_names), replacing student and tutor names with `[student]` and `[tutor]`, respectively. Models are trained on text with the structure
+```
+"[PRETEXT] {context} [TEXT] {target}"
+```
+Tutor CoPilot models are only trained with tutor utterances as targets. A synthetic example of this structure, without the full 10 context utterances, is below.
+```
+[PRETEXT] tutor: hello, [student], happy to work with you today.
+student: hi
+tutor: today we will work on the topic "adding numbers"
+...
+student: the answer is 2. [TEXT] tutor: that's correct! 2 points.
+```
 
 
 ## Model Details