Update README.md

README.md

@@ -7,7 +7,7 @@ Each folder contains a transformer model that can predict addition questions, su
 
 The folder name (e.g. sub_d6_l2_h3_t20K_s173289) contains:
 - "add", "sub", or "mix": Shows the types of questions the model can predict.
-- "d5" to "d20"" How many digits the model handles e.g. a d5 sub model can predict the answer in 123450-345670=-0123230
+- "d5" to "d20": How many digits the model handles e.g. a d5 sub model can predict the answer in 123450-345670=-0123230
 - "l1", "l2" or "l3": The number of layers in the model
 - "h3" or "h4": The number of attention heads in the model
 - "t15K" to "t85K" etc: The number of batches the model was trained on
@@ -21,9 +21,9 @@ Some folder names also contain:
 Each folder contains:
 - model.pth: The transformer model as described above
 - training_loss.json: Data gathered during model training. Used to plot "loss over training batches" graphs
-- behaviors.json: Data gathered about the behavior of the model by direct inspection. 
-- features.json: Data gathered about hypothesised algorithm features via experimentation.
+- behaviors.json: Facts gathered about the behavior of the model by direct inspection. Includes attention pattern data, PCA data, answer digit impact data, etc. 
+- features.json: Facts gathered about hypothesised algorithm features via experimentation e.g. node P12L0H1 implements the feature A3.ST. 
 
 The first 2 files were created by the https://github.com/PhilipQuirke/quanta_maths/blob/main/notebooks/VerifiedArithmeticTrain.ipynb notebook.
 The last 2 files were created by the https://github.com/PhilipQuirke/quanta_maths/blob/main/notebooks/VerifiedArithmeticAnalyse.ipynb notebook.
-
+The json file are used by the algorithm testing notebook https://github.com/PhilipQuirke/quanta_maths/blob/main/notebooks/VerifiedArithmeticAlgorithm.ipynb notebook.