Coding
PromptBeginner5 minmarkdown
Nano Banana Pro
Agent skill for nano-banana-pro
6
generic skill
Sign in to like and favorite skills
fim_rlvr_lean4/ (Lean verifier, curriculum logic, masking utilities).train_gspo_fim_20b.py, train_gspo_fim_120b.py, train_gspo_fim_local.py.data_pipeline/, dataset_analysis.md, dataset.md.test/ for unit tests; additional smoke scripts in data_pipeline/.benchmarks/verification_performance_optimization/ for verification throughput checks.verification_env/ (expects Lean toolchain accessible via lake/elan).pip install -r requirements.txt.python -m pytest test. Focused runs: python -m pytest test/test_masking.py.python -m pytest.python train_gspo_fim_20b.py (assumes data at data/fim_fresh.jsonl and Lean env ready).python train_gspo_fim_120b.py or python train_gspo_fim_local.py with matching hardware/config.test/ and follow test_*.py naming.test/; heavier smoke/benchmark scripts live outside or behind if __name__ == "__main__":.unsloth, trl, datasets) to keep tests lightweight.python -m pytest path/to/test_file.py -k pattern for targeted debugging.lake via elan; ensure ~/.elan/bin is on PATH.outputs_fim_grpo/; avoid committing large artifacts or datasets.datasets/Polars/Unsloth may require full approval (disabling the default sandbox) so /dev/shm is writable; otherwise Intel OMP SHM errors can occur. In standard Docker hosts, use --shm-size or --ipc=host as an alternative.