Datasets:
Modalities:
Text
Formats:
parquet
Size:
< 1K
Tags:
degen_test_1_countdown
budget-65k
iterative
hosted_vllm-Qwen-Qwen3-4B-Instruct-2507<api_base>http:--10.32.37.5:14280-v1
inference-engine
License:
question stringlengths 152 158 | metadata stringlengths 160 172 | task_source stringclasses 1
value | formatted_prompt listlengths 2 2 | responses_by_sample listlengths 1 1 | tokens_by_sample listlengths 1 1 | num_rounds_by_sample listlengths 1 1 | continuation_prompts_by_sample listlengths 1 1 | continuation_text_by_sample listlengths 1 1 | strategy_type stringclasses 1
value | termination_reason_by_sample listlengths 1 1 | correct_round_by_sample listlengths 1 1 | smart_termination_enabled bool 1
class | retry_count_by_sample listlengths 1 1 | request_payloads_by_sample listlengths 1 1 | answers_by_sample listlengths 1 1 | evals_by_sample listlengths 1 1 | eval_metadata_by_sample listlengths 1 1 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
"Using ALL of the numbers 12, 4, 7, 1, 2, 5, 21, 88, create an expression that equals 109. You must (...TRUNCATED) | "{\"numbers\": [12, 4, 7, 1, 2, 5, 21, 88], \"target\": 109, \"operators_used\": [\"/\", \"+\", \"/\(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**12, 4, 7, 1, 2, 5, 21, 88** \nWe must use **each number exactly o(...TRUNCATED) | [
65549
] | [
201
] | [["Think about it differently. Try a new approach to solve this problem.","Think about it differentl(...TRUNCATED) | [["We are given the numbers: \n**12, 4, 7, 1, 2, 5, 21, 88** \nWe must use **each number exactly o(...TRUNCATED) | iterative | [
null
] | [
null
] | false | [
0
] | [
[]
] | [[["88 + 21 + (5 × 2) - (4 + 1) - (12 - 7)"],["88 + 21 + (5 × 2) - (4 + 1) - (12 - 7)"],["88 + 12 (...TRUNCATED) | [[[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],(...TRUNCATED) | [[[{"answer_block":"88 + 21 + (5 * 2) - (4 + 1) - (12 - 7)","error":null,"final_answer":109.0,"is_co(...TRUNCATED) |
"Using ALL of the numbers 25, 69, 58, 14, 55, 24, 36, 60, create an expression that equals 176. You (...TRUNCATED) | "{\"numbers\": [25, 69, 58, 14, 55, 24, 36, 60], \"target\": 176, \"operators_used\": [\"-\", \"+\",(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers:\n\n**25, 69, 58, 14, 55, 24, 36, 60**\n\nWe must use **all of them exac(...TRUNCATED) | [
117265
] | [
1
] | [
[]
] | [["We are given the numbers:\n\n**25, 69, 58, 14, 55, 24, 36, 60**\n\nWe must use **all of them exac(...TRUNCATED) | iterative | [
null
] | [
null
] | false | [
0
] | [
[]
] | [[["69 + 58 + 25 + 24 + (60 - 55) - (36 - 14) = 176","69 + 58 + 25 + 24 + (14 - 14) + (55 - 55) + (3(...TRUNCATED) | [[[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,f(...TRUNCATED) | [[[{"answer_block":"69 + 58 + 25 + 24 + (60 - 55) - (36 - 14) = 176","error":"Expression does not eq(...TRUNCATED) |
"Using ALL of the numbers 55, 36, 6, 3, 14, 99, 17, 82, create an expression that equals 245. You mu(...TRUNCATED) | "{\"numbers\": [55, 36, 6, 3, 14, 99, 17, 82], \"target\": 245, \"operators_used\": [\"*\", \"-\", \(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers:\n\n**55, 36, 6, 3, 14, 99, 17, 82**\n\nWe must use **each number exactl(...TRUNCATED) | [
115965
] | [
1
] | [
[]
] | [["We are given the numbers:\n\n**55, 36, 6, 3, 14, 99, 17, 82**\n\nWe must use **each number exactl(...TRUNCATED) | iterative | [
null
] | [
null
] | false | [
0
] | [
[]
] | [[["82 + 99 + 55 + (36 / 6) + (17 - 14) + 3 - 3","82 + 99 + 55 + 36 / 6 + 17 - 14 + 3 - 3","82 + 99 (...TRUNCATED) | [[[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,f(...TRUNCATED) | [[[{"answer_block":"82 + 99 + 55 + (36 / 6) + (17 - 14) + 3 - 3","error":"Expression uses an unavail(...TRUNCATED) |
"Using ALL of the numbers 87, 87, 29, 2, 19, 4, 1, 1, create an expression that equals 281. You must(...TRUNCATED) | "{\"numbers\": [87, 87, 29, 2, 19, 4, 1, 1], \"target\": 281, \"operators_used\": [\"/\", \"+\", \"/(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**87, 87, 29, 2, 19, 4, 1, 1** \nWe must use **each number exactly (...TRUNCATED) | [
65558
] | [
278
] | [["Think about it differently. Try a new approach to solve this problem.","Think about it differentl(...TRUNCATED) | [["We are given the numbers: \n**87, 87, 29, 2, 19, 4, 1, 1** \nWe must use **each number exactly (...TRUNCATED) | iterative | [
"budget_reached"
] | [
null
] | false | [
0
] | [[{"endpoint_type":"completions","litellm_function":"atext_completion","raw_prompt":"<|im_start|>sys(...TRUNCATED) | [[["(87 + 87) + (4 × 19) + 29 + 2 + (1 - 1) = 281"],["(87 + 87) + (19 × 4) + 29 + 2 + (1 - 1) = 28(...TRUNCATED) | [[[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],(...TRUNCATED) | [[[{"answer_block":"(87 + 87) + (4 * 19) + 29 + 2 + (1 - 1) = 281","error":null,"final_answer":281.0(...TRUNCATED) |
"Using ALL of the numbers 56, 63, 7, 50, 44, 1, 87, 52, create an expression that equals 132. You mu(...TRUNCATED) | "{\"numbers\": [56, 63, 7, 50, 44, 1, 87, 52], \"target\": 132, \"operators_used\": [\"-\", \"/\", \(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**56, 63, 7, 50, 44, 1, 87, 52** \nWe must use **all of them exactl(...TRUNCATED) | [
65560
] | [
255
] | [["Think about it differently. Try a new approach to solve this problem.","Think about it differentl(...TRUNCATED) | [["We are given the numbers: \n**56, 63, 7, 50, 44, 1, 87, 52** \nWe must use **all of them exactl(...TRUNCATED) | iterative | [
"budget_reached"
] | [
null
] | false | [
0
] | [[{"endpoint_type":"completions","litellm_function":"atext_completion","raw_prompt":"<|im_start|>sys(...TRUNCATED) | [[["(87 + 52 - 7) + ((63 - 56) - (50 - 44) - 1)"],["(87 + 52 - 7) + ((63 - 56) - (50 - 44) - 1)"],["(...TRUNCATED) | [[[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],[true],(...TRUNCATED) | [[[{"answer_block":"(87 + 52 - 7) + ((63 - 56) - (50 - 44) - 1)","error":null,"final_answer":132.0,"(...TRUNCATED) |
"Using ALL of the numbers 7, 1, 49, 7, 28, 59, 3, 44, create an expression that equals 521. You must(...TRUNCATED) | "{\"numbers\": [7, 1, 49, 7, 28, 59, 3, 44], \"target\": 521, \"operators_used\": [\"/\", \"*\", \"/(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**7, 1, 49, 7, 28, 59, 3, 44** \nWe must use **each number exactly (...TRUNCATED) | [
65543
] | [
197
] | [["Think about it differently. Try a new approach to solve this problem.","Think about it differentl(...TRUNCATED) | [["We are given the numbers: \n**7, 1, 49, 7, 28, 59, 3, 44** \nWe must use **each number exactly (...TRUNCATED) | iterative | [
null
] | [
null
] | false | [
0
] | [
[]
] | [[["44 × (59 - 49 + 3 - 1) - 7 + (7 - 7) × 28"],["49 × (7 + 3) + (59 - 28) + (7 - 7) × (44 - 1)"(...TRUNCATED) | [[[false],[false],[false],[false],[false],[false],[false],[false],[false],[false],[false],[false],[f(...TRUNCATED) | [[[{"answer_block":"44 * (59 - 49 + 3 - 1) - 7 + (7 - 7) * 28","error":"Expression uses an unavailab(...TRUNCATED) |
"Using ALL of the numbers 97, 4, 93, 26, 3, 24, 20, 31, create an expression that equals 620. You mu(...TRUNCATED) | "{\"numbers\": [97, 4, 93, 26, 3, 24, 20, 31], \"target\": 620, \"operators_used\": [\"-\", \"/\", \(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**97, 4, 93, 26, 3, 24, 20, 31** \nWe must use **each number exactl(...TRUNCATED) | [
114653
] | [
1
] | [
[]
] | [["We are given the numbers: \n**97, 4, 93, 26, 3, 24, 20, 31** \nWe must use **each number exactl(...TRUNCATED) | iterative | [
null
] | [
null
] | false | [
0
] | [
[]
] | [[["31 * 20 + (97 - 93) - (26 - 24) - (4 - 3)","(97 + 93) * 3 + 26 + 24 + 4 - 20 - 31","31 * 20 + (9(...TRUNCATED) | [[[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,f(...TRUNCATED) | [[[{"answer_block":"31 * 20 + (97 - 93) - (26 - 24) - (4 - 3)","error":"Expression does not equal st(...TRUNCATED) |
"Using ALL of the numbers 45, 69, 55, 40, 1, 9, 8, 83, create an expression that equals 913. You mus(...TRUNCATED) | "{\"numbers\": [45, 69, 55, 40, 1, 9, 8, 83], \"target\": 913, \"operators_used\": [\"*\", \"+\", \"(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**45, 69, 55, 40, 1, 9, 8, 83** \nWe must use **each number exactly(...TRUNCATED) | [
119055
] | [
1
] | [
[]
] | [["We are given the numbers: \n**45, 69, 55, 40, 1, 9, 8, 83** \nWe must use **each number exactly(...TRUNCATED) | iterative | [
null
] | [
null
] | false | [
0
] | [
[]
] | [[["83 × (55 - 45 + 1) + (69 - 69) + (40 - 40) + (9 - 9)","83 × (55 - 45 + 1) = 913","83 * (55 - 4(...TRUNCATED) | [[[false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,false,f(...TRUNCATED) | [[[{"answer_block":"83 * (55 - 45 + 1) + (69 - 69) + (40 - 40) + (9 - 9)","error":"Expression uses a(...TRUNCATED) |
"Using ALL of the numbers 72, 3, 67, 1, 43, 1, 93, 53, create an expression that equals 145. You mus(...TRUNCATED) | "{\"numbers\": [72, 3, 67, 1, 43, 1, 93, 53], \"target\": 145, \"operators_used\": [\"/\", \"-\", \"(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**72, 3, 67, 1, 43, 1, 93, 53** \nWe must use **each number exactly(...TRUNCATED) | [
65537
] | [
221
] | [["Think about it differently. Try a new approach to solve this problem.","Think about it differentl(...TRUNCATED) | [["We are given the numbers: \n**72, 3, 67, 1, 43, 1, 93, 53** \nWe must use **each number exactly(...TRUNCATED) | iterative | [
"budget_reached"
] | [
null
] | false | [
0
] | [[{"endpoint_type":"completions","litellm_function":"atext_completion","raw_prompt":"<|im_start|>sys(...TRUNCATED) | [[["72 + 43 + 3 + (93 - 67) + 1 + (53 / 53) - 1"],["93 + 43 + (72 - 67) + 3 + 1 + (53 / 53) - 1"],["(...TRUNCATED) | [[[false],[false],[false],[false],[false],[false,false,false],[false],[false],[false],[false],[false(...TRUNCATED) | [[[{"answer_block":"72 + 43 + 3 + (93 - 67) + 1 + (53 / 53) - 1","error":"Expression uses an unavail(...TRUNCATED) |
"Using ALL of the numbers 13, 1, 13, 68, 2, 2, 93, 62, create an expression that equals 102. You mus(...TRUNCATED) | "{\"numbers\": [13, 1, 13, 68, 2, 2, 93, 62], \"target\": 102, \"operators_used\": [\"/\", \"-\", \"(...TRUNCATED) | countdown | [{"content":"You are a helpful assistant that solves math problems by reasoning step-by-step. Show y(...TRUNCATED) | [["We are given the numbers: \n**13, 1, 13, 68, 2, 2, 93, 62** \nWe must use **each number exactly(...TRUNCATED) | [
65544
] | [
259
] | [["Think about it differently. Try a new approach to solve this problem.","Think about it differentl(...TRUNCATED) | [["We are given the numbers: \n**13, 1, 13, 68, 2, 2, 93, 62** \nWe must use **each number exactly(...TRUNCATED) | iterative | [
"budget_reached"
] | [
null
] | false | [
0
] | [[{"endpoint_type":"completions","litellm_function":"atext_completion","raw_prompt":"<|im_start|>sys(...TRUNCATED) | [[["93 + 68 - 62 + 13 - 13 + 2 + 2 - 1"],["(68 + 62) - (13 + 13) - (2 / 2) - 1"],["(68 + 62) - (13 +(...TRUNCATED) | [[[true],[false],[false],[false],[false],[false],[false],[false],[false],[false],[false],[false],[fa(...TRUNCATED) | [[[{"answer_block":"93 + 68 - 62 + 13 - 13 + 2 + 2 - 1","error":null,"final_answer":102.0,"is_correc(...TRUNCATED) |
End of preview. Expand in Data Studio
PA-Qwen3-4B-Instruct-2507-cd8arg-default-iter-64k
Countdown reasoning with iterative strategy to 65536 tokens
Dataset Info
- Rows: 100
- Columns: 18
Columns
| Column | Type | Description |
|---|---|---|
| question | Value('string') | The countdown problem statement |
| metadata | Value('string') | JSON with target number, available numbers, and solution |
| task_source | Value('string') | Task type identifier (countdown) |
| formatted_prompt | List({'content': Value('string'), 'role': Value('string')}) | Initial prompt messages sent to the model |
| responses_by_sample | List(List(Value('string'))) | Full model response [sample_idx][round_idx] = response string |
| tokens_by_sample | List(Value('int64')) | Total token count for each sample (based on continuation_text) |
| num_rounds_by_sample | List(Value('int64')) | Number of continuation rounds for each sample |
| continuation_prompts_by_sample | List(List(Value('string'))) | Continuation prompts used [sample_idx][round_idx] = prompt string |
| continuation_text_by_sample | List(List(Value('string'))) | Truncated text used for continuation [sample_idx][round_idx]. For budget_forcing: truncated at first tag. For other strategies: same as responses_by_sample. |
| strategy_type | Value('string') | Prompting strategy: 'iterative', 'budget_forcing', or 'random_noise' |
| termination_reason_by_sample | List(Value('string')) | Why each sample terminated: 'max_rounds_reached', 'budget_reached', or 'answer_correct' |
| correct_round_by_sample | List(Value('null')) | Round number where correct answer was first found (None if never correct) |
| smart_termination_enabled | Value('bool') | Whether smart termination was enabled for this run |
| retry_count_by_sample | List(Value('int64')) | Total retry attempts per sample due to empty or too-short responses (0 if no retries needed) |
| request_payloads_by_sample | List(List({'endpoint_type': Value('string'), 'litellm_function': Value('string'), 'raw_prompt': Value('string'), 'raw_system_prompt': Value('null'), 'request_payload': {'api_base': Value('string'), 'api_key': Value('string'), 'max_tokens': Value('int64'), 'model': Value('string'), 'prompt': Value('string'), 'stream': Value('bool'), 'temperature': Value('float64'), 'timeout': Value('int64')}})) | Raw LiteLLM request payloads [sample_idx][round_idx] = payload dict (only when --use-inference-engine) |
| answers_by_sample | List(List(List(Value('string')))) | Extracted answer spans [sample][round][answer_idx] |
| evals_by_sample | List(List(List(Value('bool')))) | Correctness evaluations [sample][round][answer_idx] |
| eval_metadata_by_sample | List(List(List({'answer_block': Value('string'), 'error': Value('string'), 'final_answer': Value('float64'), 'is_correct': Value('bool')}))) | Evaluation details [sample][round][answer_idx] |
Generation Parameters
{
"script_name": "stage_01_generate_reasoning.py",
"model": "hosted_vllm/Qwen/Qwen3-4B-Instruct-2507<api_base>http://10.32.37.5:14280/v1",
"hyperparameters": {
"temperature": 0.7,
"max_tokens_per_round": 65536,
"target_total_tokens": 65536,
"samples": 1,
"strategy": "iterative",
"max_rounds": null,
"min_tokens": null,
"max_retries_per_round": -1,
"max_total_tokens": 68000
},
"input_datasets": [],
"description": "Countdown reasoning with iterative strategy to 65536 tokens",
"experiment": "degen_test_1_countdown",
"strategy": "iterative",
"arg_count": 8,
"seed": 42,
"smart_termination": false,
"max_rounds": null,
"min_tokens": null,
"inference_backend": "inference_engine",
"prompt_config": "default",
"enable_thinking": false,
"status": "completed",
"completed_at": "2026-01-18T21:13:33.709780"
}
Experiment Documentation
For complete experiment details, see https://github.com/TAUR-Lab/Reasoning-Horizons/tree/main/experiments/degen_test_1_countdown
Usage
from datasets import load_dataset
dataset = load_dataset("reasoning-degeneration-dev/PA-Qwen3-4B-Instruct-2507-cd8arg-default-iter-64k", split="train")
print(f"Loaded {len(dataset)} rows")
This dataset is tracked in reasoning-degeneration-dev/PROJECT-MANIFEST
- Downloads last month
- 7