My take on improving LLM4Decompile at decomp through harvest output pairings to randomized inputs.
Python
0
21 commits
updated Sep 14, 2026
Research into training LLM4Decompile to recover runnable C from compiled binaries using GRPO and execution-based rewards on randomized ExeBench inputs.
The training prototype is in train.py; the isolated verifier and data pipeline are still in progress. See .agent/checklist.md for the research plan.
My take on improving LLM4Decompile at decomp through harvest output pairings to randomized inputs.
Python
0
21 commits
updated Sep 14, 2026
Research into training LLM4Decompile to recover runnable C from compiled binaries using GRPO and execution-based rewards on randomized ExeBench inputs.
The training prototype is in train.py; the isolated verifier and data pipeline are still in progress. See .agent/checklist.md for the research plan.
