question string | metadata string | task_source string | formatted_prompt list | response list | formatted_prompt__metadata list | response__metadata list | response__extractions list | response__extractions_metadata list | eval_correct list | eval_correct__metadata list | _judge_prompt__metadata list | _judge_response__metadata list | strategy_compliance int64 | compliance_reasoning string | compliance_elements_followed list | compliance_elements_missed list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
"Using ALL of the numbers 28, 98, 44, 14, 69, 49, 7, 46, create an expression that equals 303. You m(...TRUNCATED) | "{\"numbers\": [28, 98, 44, 14, 69, 49, 7, 46], \"target\": 303, \"operators_used\": [\"*\", \"*\", (...TRUNCATED) | countdown | [{"content":"You are a math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n28, 98, 44, 14, 69, 49, 7, 46\n\nAnd we need to use **all** of them *(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.049309,"finish_reason":"length","input_tokens":157,"model":"together_ai/moonshotai/Kimi-K(...TRUNCATED) | [
[]
] | [
[]
] | [
[]
] | [
[]
] | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0026690999999999998,"finish_reason":"stop","input_tokens":16994,"model":"gpt-4o-mini-2024(...TRUNCATED) | 1 | "The model completely ignores the prescribed strategy and relies on ad-hoc reasoning and estimation (...TRUNCATED) | [] | ["Lists all available numbers and target at the start (INVENTORY step)","Picks pairs of numbers and (...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 math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n25, 69, 58, 14, 55, 24, 36, 60\n\nWe need to use **all of them exactl(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.049309,"finish_reason":"length","input_tokens":157,"model":"together_ai/moonshotai/Kimi-K(...TRUNCATED) | [
[]
] | [
[]
] | [
[]
] | [
[]
] | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0022797,"finish_reason":"stop","input_tokens":14370,"model":"gpt-4o-mini-2024-07-18","out(...TRUNCATED) | 1 | "The model completely ignores the systematic tree search strategy and instead relies on ad-hoc reaso(...TRUNCATED) | [] | ["Lists all available numbers and target at the start (INVENTORY step)","Picks pairs of numbers and (...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 math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n**55, 36, 6, 3, 14, 99, 17, 82**\n\nWe must use **each number exactly(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.027142000000000003,"finish_reason":"stop","input_tokens":157,"model":"together_ai/moonsho(...TRUNCATED) | [
[
"99 + 82 + 55 + (36 / (17 - 14) - (6 - 3)) = 245"
]
] | [
[
{
"end_index": 23326,
"end_phrase": "</answer>",
"start_index": 23262,
"start_phrase": "<answer>"
}
]
] | [
[
true
]
] | [[{"answer_block":"99 + 82 + 55 + (36 / (17 - 14) - (6 - 3)) = 245","error":null,"final_answer":245.(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0015657,"finish_reason":"stop","input_tokens":9766,"model":"gpt-4o-mini-2024-07-18","outp(...TRUNCATED) | 2 | "The model acknowledges the need to use all numbers exactly once and attempts to find a solution, bu(...TRUNCATED) | ["Lists all available numbers and target at the start","Uses some operations to try to reach the tar(...TRUNCATED) | ["Picks pairs of numbers and tries operations systematically","Tracks the available number set after(...TRUNCATED) |
"Using ALL of the numbers 87, 69, 43, 4, 55, 34, 23, 2, create an expression that equals 874. You mu(...TRUNCATED) | "{\"numbers\": [87, 69, 43, 4, 55, 34, 23, 2], \"target\": 874, \"operators_used\": [\"*\", \"+\", \(...TRUNCATED) | countdown | [{"content":"You are a math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n87, 69, 43, 4, 55, 34, 23, 2\n\nAnd we need to use **all of them exac(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.049309,"finish_reason":"length","input_tokens":157,"model":"together_ai/moonshotai/Kimi-K(...TRUNCATED) | [
[]
] | [
[]
] | [
[]
] | [
[]
] | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0027561,"finish_reason":"stop","input_tokens":17562,"model":"gpt-4o-mini-2024-07-18","out(...TRUNCATED) | 1 | "The model completely ignores the systematic tree search strategy and relies on ad-hoc reasoning and(...TRUNCATED) | [] | ["Lists all available numbers and target at the start (INVENTORY step)","Picks pairs of numbers and (...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 math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n56, 63, 7, 50, 44, 1, 87, 52\n\nWe need to form an expression using *(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.011916999999999999,"finish_reason":"stop","input_tokens":157,"model":"together_ai/moonsho(...TRUNCATED) | [
[
"87 + 44 + 1 + (63 - 7 - 56) * (50 + 52) = 132"
]
] | [
[
{
"end_index": 11331,
"end_phrase": "</answer>",
"start_index": 11269,
"start_phrase": "<answer>"
}
]
] | [
[
true
]
] | [[{"answer_block":"87 + 44 + 1 + (63 - 7 - 56) * (50 + 52) = 132","error":null,"final_answer":132.0,(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.00085095,"finish_reason":"stop","input_tokens":5057,"model":"gpt-4o-mini-2024-07-18","out(...TRUNCATED) | 2 | "The model acknowledges the need to use all numbers and operations but quickly deviates into heurist(...TRUNCATED) | [
"Lists all available numbers and target at the start"
] | ["Picks pairs of numbers and tries operations systematically","Tracks the available number set after(...TRUNCATED) |
"Using ALL of the numbers 32, 32, 36, 58, 2, 97, 60, 11, create an expression that equals 409. You m(...TRUNCATED) | "{\"numbers\": [32, 32, 36, 58, 2, 97, 60, 11], \"target\": 409, \"operators_used\": [\"/\", \"*\", (...TRUNCATED) | countdown | [{"content":"You are a math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n32, 32, 36, 58, 2, 97, 60, 11\n\nWe need to use **each number exactly(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.01069,"finish_reason":"stop","input_tokens":157,"model":"together_ai/moonshotai/Kimi-K2-I(...TRUNCATED) | [
[
"97 × ((36 − 32) + (60 − 58 − 2)) + (32 − 11) = 409"
]
] | [
[
{
"end_index": 9133,
"end_phrase": "</answer>",
"start_index": 9066,
"start_phrase": "<answer>"
}
]
] | [
[
false
]
] | [[{"answer_block":"97 * ((36 − 32) + (60 − 58 − 2)) + (32 − 11) = 409","error":"Could not ev(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.00079065,"finish_reason":"stop","input_tokens":4647,"model":"gpt-4o-mini-2024-07-18","out(...TRUNCATED) | 2 | "The model acknowledges the need to use all numbers and attempts to find combinations, but it primar(...TRUNCATED) | ["Lists all available numbers and target at the start","Picks pairs of numbers and tries operations (...TRUNCATED) | ["Tracks the available number set after each operation","Uses explicit tree structure","Explicitly m(...TRUNCATED) |
"Using ALL of the numbers 38, 30, 47, 23, 39, 2, 91, 1, create an expression that equals 785. You mu(...TRUNCATED) | "{\"numbers\": [38, 30, 47, 23, 39, 2, 91, 1], \"target\": 785, \"operators_used\": [\"-\", \"*\", \(...TRUNCATED) | countdown | [{"content":"You are a math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers: \n38, 30, 47, 23, 39, 2, 91, 1 \nAnd we need to create an expression u(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.016411000000000002,"finish_reason":"stop","input_tokens":157,"model":"together_ai/moonsho(...TRUNCATED) | [
[
"91 * (47 - 38) - (39 + 2 - 30 + 23) * 1 = 785"
]
] | [
[
{
"end_index": 12820,
"end_phrase": "</answer>",
"start_index": 12758,
"start_phrase": "<answer>"
}
]
] | [
[
true
]
] | [[{"answer_block":"91 * (47 - 38) - (39 + 2 - 30 + 23) * 1 = 785","error":null,"final_answer":785.0,(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0011060999999999998,"finish_reason":"stop","input_tokens":6558,"model":"gpt-4o-mini-2024-(...TRUNCATED) | 1 | "The model completely ignores the systematic tree search strategy and relies on ad-hoc reasoning and(...TRUNCATED) | [] | ["Lists all available numbers and target at the start (INVENTORY step)","Picks pairs of numbers and (...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 math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n13, 1, 13, 68, 2, 2, 93, 62\n\nWe need to use **all of them exactly o(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.035325999999999996,"finish_reason":"stop","input_tokens":157,"model":"together_ai/moonsho(...TRUNCATED) | [
[
"(2 × 62 - 93 + 68 + 1 + 2) × (13 / 13) = 102"
]
] | [
[
{
"end_index": 29873,
"end_phrase": "</answer>",
"start_index": 29812,
"start_phrase": "<answer>"
}
]
] | [
[
true
]
] | [[{"answer_block":"(2 * 62 - 93 + 68 + 1 + 2) * (13 / 13) = 102","error":null,"final_answer":102.0,"(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0020511,"finish_reason":"stop","input_tokens":12886,"model":"gpt-4o-mini-2024-07-18","out(...TRUNCATED) | 2 | "The model acknowledges the need to use all numbers and the target but quickly reverts to a more heu(...TRUNCATED) | [
"Lists all available numbers and target at the start (INVENTORY step)"
] | ["Picks pairs of numbers and tries operations systematically (BRANCH step)","Tracks the available nu(...TRUNCATED) |
"Using ALL of the numbers 87, 1, 21, 9, 57, 45, 6, 17, create an expression that equals 113. You mus(...TRUNCATED) | "{\"numbers\": [87, 1, 21, 9, 57, 45, 6, 17], \"target\": 113, \"operators_used\": [\"/\", \"*\", \"(...TRUNCATED) | countdown | [{"content":"You are a math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n87, 1, 21, 9, 57, 45, 6, 17\n\nWe must use **all** of them exactly on(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.049309,"finish_reason":"length","input_tokens":157,"model":"together_ai/moonshotai/Kimi-K(...TRUNCATED) | [
[]
] | [
[]
] | [
[]
] | [
[]
] | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0027534,"finish_reason":"stop","input_tokens":17528,"model":"gpt-4o-mini-2024-07-18","out(...TRUNCATED) | 1 | "The model completely ignores the systematic tree search strategy and instead relies on a more intui(...TRUNCATED) | [] | ["Lists all available numbers and target at the start (INVENTORY step)","Picks pairs of numbers and (...TRUNCATED) |
"Using ALL of the numbers 60, 90, 39, 30, 53, 15, 18, 6, create an expression that equals 239. You m(...TRUNCATED) | "{\"numbers\": [60, 90, 39, 30, 53, 15, 18, 6], \"target\": 239, \"operators_used\": [\"+\", \"*\", (...TRUNCATED) | countdown | [{"content":"You are a math problem solver. Solve the given arithmetic problem step by step. Show yo(...TRUNCATED) | ["We are given the numbers:\n\n60, 90, 39, 30, 53, 15, 18, 6\n\nWe need to use **all of them exactly(...TRUNCATED) | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"together_ai/moonshotai/Kimi-K2-In(...TRUNCATED) | [{"cost":0.049309,"finish_reason":"length","input_tokens":157,"model":"together_ai/moonshotai/Kimi-K(...TRUNCATED) | [
[]
] | [
[]
] | [
[]
] | [
[]
] | [{"endpoint_type":"chat","litellm_function":"acompletion","model":"gpt-4o-mini-2024-07-18","raw_prom(...TRUNCATED) | [{"cost":0.0026006999999999996,"finish_reason":"stop","input_tokens":16534,"model":"gpt-4o-mini-2024(...TRUNCATED) | 1 | "The model completely disregards the prescribed strategy and relies on ad-hoc reasoning and estimati(...TRUNCATED) | [] | ["Lists all available numbers and target at the start (INVENTORY step)","Picks pairs of numbers and (...TRUNCATED) |
t1-strategy-countdown-baseline-kimi-k2-instruct-kimi-inst
Strategy compliance evaluation on countdown arithmetic — baseline variant.
Model received no strategy instruction (baseline control). Judge still scores against tree search rubric. Compliance is scored by an LLM judge (1-5 Likert).
Results
| Metric | Value |
|---|---|
| pass@1 | 0.4000 |
| Strategy compliance (mean) | 1.40 |
| Strategy compliance (min) | 1 |
| Strategy compliance (max) | 2 |
| Total problems | 10 |
Compliance Distribution
| Score | Count | Meaning |
|---|---|---|
| 1 | 6 | Non-compliant |
| 2 | 4 | Minimal |
| 3 | 0 | Partial |
| 4 | 0 | Mostly compliant |
| 5 | 0 | Fully compliant |
Details
| Parameter | Value |
|---|---|
| Model | together_ai/moonshotai/Kimi-K2-Instruct-0905 |
| Thinking | False |
| Temperature | 0.7 |
| Max tokens | 32768 |
| Judge model | openai/gpt-4o-mini |
| Num examples | 10 |
| Num args | 8 |
| Variant | baseline |
Usage
from datasets import load_dataset
ds = load_dataset("reasoning-degeneration-dev/t1-strategy-countdown-baseline-kimi-k2-instruct-kimi-inst", split="train")
scores = ds["strategy_compliance"]
print(f"Mean compliance: {sum(scores) / len(scores):.2f}")
Tracked in reasoning-degeneration-dev/PROJECT-MANIFEST
- Downloads last month
- 11