metadata
license: mit
tags:
- multiplex-thinking
- countdown
- input-data
- eval
multiplex-countdown-eval-data-v1
Countdown d4-d5 evaluation data for multiplex-thinking-countdown experiment. 500 held-out problems.
Dataset Info
- Rows: 500
- Columns: 5
Columns
| Column | Type | Description |
|---|---|---|
| data_source | Value('large_string') | Task source identifier (countdown) |
| prompt | List({'content': Value('string'), 'role': Value('string')}) | Chat-formatted prompt with system + user messages |
| ability | Value('large_string') | Task ability type |
| reward_model | {'ground_truth': Value('string'), 'style': Value('string')} | Ground truth info for reward calculation |
| extra_info | {'difficulty': Value('int64'), 'index': Value('int64'), 'numbers': List(Value('int64')), 'split': Value('string'), 'target': Value('int64')} | Additional metadata (target number, available numbers) |
Generation Parameters
{
"script_name": "generate_countdown_data.py",
"description": "Countdown d4-d5 evaluation data for multiplex-thinking-countdown experiment. 500 held-out problems.",
"model": "deepseek-ai/DeepSeek-R1-Distill-Qwen-7B",
"hyperparameters": {},
"input_datasets": []
}
Experiment Documentation
For complete experiment details, see https://github.com/Zayne-sprague/SC-Research-Notes/tree/main/experiments/multiplex-thinking-countdown
Usage
from datasets import load_dataset
dataset = load_dataset("reasoning-degeneration-dev/multiplex-countdown-eval-data-v1", split="train")
print(f"Loaded {len(dataset)} rows")
This dataset is tracked in reasoning-degeneration-dev/PROJECT-MANIFEST