AI_Fiends_4_This
Execution-verified repair trajectory dataset developed in collaboration with WithIn Us AI.
Overview
AI_Fiends_4_This is a large-scale execution-grounded synthetic repair dataset focused on:
- autonomous debugging
- repair supervision
- execution-aware reasoning
- traceback interpretation
- assertion verification
- process-supervised fine-tuning
Dataset Statistics
- Total Rows: 50000
- Domains: math, code, science, reasoning, algorithms, data_processing, automation, debugging
- Bug Types: syntax, name_error, type_error, logic, assertion, index_error, key_error, zero_division
- Repair Success Rate: 100.0%
- Execution Verified: Yes
- Assertions Included: Yes
- Deduplicated: Yes
Schema
- prompt
- domain
- bug_type
- difficulty
- code_v1
- stdout_v1
- stderr_v1
- repair_reasoning
- code_v2
- execution_output
- stderr_v2
- repair_verified
- runtime_ms
- traceback_chars
Focus
This dataset is designed for:
- execution-aware LLMs
- autonomous coding agents
- repair-focused SFT
- debugging systems
- process reward modeling
- recursive execution training
Developed By
WithIn Us AI