Jongbin-kr/VeriReason-reasoning-reproduced-1513_luna-medium
VeriReason reasoning reproduced (Luna medium) This dataset is a deterministic sample of the locally updated reproduced VeriReason files. Source files: train (1).jsonl, validation (1).jsonl Sampling: independent shuffle then prefix slice, fixed seed 42 Splits: train 1513, validation 189 Original target dataset: Jongbin-kr/VeriReason-RTL-Coder_7b_reasoning_tb (train 1513 / validation 189) Schema: id, instruction, output, tb, tb_result Quality checks: valid JSON, unique IDs… See the full description on the dataset page: https://huggingface.co/datasets/Jongbin-kr/VeriReason-reasoning-reproduced-1513_luna-medium.
049
VeriReason reasoning reproduced (Luna medium)
This dataset is a deterministic sample of the locally updated reproduced VeriReason files.
- Source files: train (1).jsonl, validation (1).jsonl
- Sampling: independent shuffle then prefix slice, fixed seed 42
- Splits: train 1513, validation 189
- Original target dataset: Jongbin-kr/VeriReason-RTL-Coder7breasoning_tb (train 1513 / validation 189)
- Schema: id, instruction, output, tb, tb_result
- Quality checks: valid JSON, unique IDs, non-overlapping splits, one ordered think/answer block per output, no explicit failure markers in tb_result.
- Normalization: removed only non-whitespace text after the final endmodule in 120 rows because it is not valid Verilog syntax.
- A Verilog simulator was not available in the preparation environment, so full recompilation was not performed; tb_result was checked for explicit FAIL/ERROR/FATAL/TIMEOUT markers and testbench boundaries were statically checked.
