Pinned Loading

1. ml-bugfix-bench ml-bugfix-bench Public

   30 original bugs. Every one fails with the buggy code and passes with the fix.

   Python

2. hybrid-rag-eval hybrid-rag-eval Public

   Retrieval over 744 Python PEPs (the Python language’s design documents) on 45 hand-written questions. Plain keyword search (BM25) gets the right PEP in its top 5 for 84% of questions and beats the …

   Python

3. sql-analyst-agent sql-analyst-agent Public

   30 questions with verified answer queries. A built-in “perfect model” check scores 100%, which proves the grader works. Real model scores need an API key.

   Python

4. lora-text-to-sql lora-text-to-sql Public

   Code and Colab notebook are ready. The results table is waiting on a GPU run.

   Python