Upload paper.md with huggingface_hub

paper.md

@@ -41,7 +41,7 @@ Our contributions are:
 ### 3.1 Dataset
 
 We use 22,452 Ruby methods extracted from open-source repositories, each parsed into an AST with:
-- **74-dimensional node features** encoding node type (one-hot), depth, sibling index, and subtree statistics.
+- **74-dimensional node features** encoding node type (one-hot over 73 known AST types plus one `UNKNOWN` token). Lexical literals — identifiers, string contents, numeric values — are stripped of their content and mapped to this single `UNKNOWN` token to bound the feature space. As we show in Section 4.4, this design choice has profound consequences for generation.
 - **Edge attributes** (3D): edge type encoding, relative depth, and child index.
 - **Positional encodings** (2D): tree depth and sibling position.
 - **Labels**: McCabe cyclomatic complexity (integer, range 1–200+).