Upload README.md with huggingface_hub

README.md

@@ -11,6 +11,7 @@ Repository-level code completion dataset for hypernetwork-based LoRA generation.
 ```
 splits/main/           # Original splits (train, cr_val, cr_test, ir_val, ir_test + structured)
 splits/expanded/       # Expanded training (~609 repos)
-oracle_context_cache/  # Pre-computed DRC context (370 MB, 512 repos)
+oracle_context_cache/   # Pre-computed DRC context v2 (uncompressed, 512 repos)
+oracle_context_cache_v3/ # Compressed DRC context v3 (6K token budget, 512 repos)
 evaluation_results/    # Pre-computed baseline JSONs
 ```