uid large_stringlengths 6 43 | source large_stringclasses 10
values | stage large_stringclasses 2
values | model large_stringclasses 10
values | hf_id large_stringclasses 10
values | revision large_stringclasses 10
values | thinking_mode large_stringclasses 3
values | solve_idx int64 -1 -1 | sample_idx int64 0 9 | seed int64 20.3M 21.3M | prompt_text large_stringlengths 314 10.5k | template_kwargs large_stringclasses 3
values | generation_prefix large_stringclasses 2
values | prompt_tokens int64 63 4.24k | max_tokens int64 28.5k 65.4k | raw_output large_stringlengths 1 358k | output large_stringlengths 1 358k | thinking_text large_stringlengths 0 283k | content_text large_stringlengths 0 358k | completion_tokens int64 2 65.4k | finish_reason large_stringclasses 2
values | truncated bool 2
classes | thinking_closed bool 2
classes | forced_status stringclasses 2
values | forced_output stringclasses 56
values | skip_reason null | conf_raw int64 0 100 ⌀ | conf int64 0 100 ⌀ | conf_off_grid bool 2
classes | conf_extract_method large_stringclasses 2
values | pre_leak bool 2
classes | leak_reasons large_stringclasses 7
values | leak_candidate large_stringlengths 1 148k ⌀ | conf_ambiguous bool 2
classes |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
aime_2025_1 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 0 | 20,831,573 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 203 | 65,301 | I will assess the problem step by step as instructed.
**1. What does the problem require?**
The problem asks for all integer bases \( b > 9 \) such that the number written as "17" in base \( b \) (i.e., the value \( 1 \times b + 7 \)) is a divisor of the number written as "97" in base \( b \) (i.e., the value \( 9 \t... | I will assess the problem step by step as instructed.
**1. What does the problem require?**
The problem asks for all integer bases \( b > 9 \) such that the number written as "17" in base \( b \) (i.e., the value \( 1 \times b + 7 \)) is a divisor of the number written as "97" in base \( b \) (i.e., the value \( 9 \t... | I will assess the problem step by step as instructed.
**1. What does the problem require?**
The problem asks for all integer bases \( b > 9 \) such that the number written as "17" in base \( b \) (i.e., the value \( 1 \times b + 7 \)) is a divisor of the number written as "97" in base \( b \) (i.e., the value \( 9 \t... | 673 | stop | false | true | null | null | null | 70 | 70 | false | rule_boxed | false | [] | null | true | ||
aime_2025_1 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 1 | 20,831,574 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 203 | 65,301 | I need to assess my ability to solve the given problem before providing an estimate.
Let's break down what the problem requires:
1. **Understanding Base Notation:** The numbers 17_b and 97_b are in base b. I must convert these to base-10 integers. For 17_b, that's 1*b + 7. For 97_b, that's 9*b + 7. This requires reco... | I need to assess my ability to solve the given problem before providing an estimate.
Let's break down what the problem requires:
1. **Understanding Base Notation:** The numbers 17_b and 97_b are in base b. I must convert these to base-10 integers. For 17_b, that's 1*b + 7. For 97_b, that's 9*b + 7. This requires reco... | I need to assess my ability to solve the given problem before providing an estimate.
Let's break down what the problem requires:
1. **Understanding Base Notation:** The numbers 17_b and 97_b are in base b. I must convert these to base-10 integers. For 17_b, that's 1*b + 7. For 97_b, that's 9*b + 7. This requires reco... | 919 | stop | false | true | null | null | null | 80 | 80 | false | rule_boxed | false | [] | null | false | ||
aime_2025_1 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 2 | 20,831,575 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 203 | 65,301 | I will assess the problem as requested.
First, I need to understand what the problem is asking. It involves numbers written in different bases—specifically, the numbers 17 and 97 in base \( b \). I must interpret these numbers in base 10, set up the divisibility condition, and then solve for integer bases greater than... | I will assess the problem as requested.
First, I need to understand what the problem is asking. It involves numbers written in different bases—specifically, the numbers 17 and 97 in base \( b \). I must interpret these numbers in base 10, set up the divisibility condition, and then solve for integer bases greater than... | I will assess the problem as requested.
First, I need to understand what the problem is asking. It involves numbers written in different bases—specifically, the numbers 17 and 97 in base \( b \). I must interpret these numbers in base 10, set up the divisibility condition, and then solve for integer bases greater than... | 429 | stop | false | true | null | null | null | 70 | 70 | false | rule_boxed | false | [] | null | true | ||
aime_2025_10 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 0 | 21,220,471 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 471 | 65,033 | To assess how likely I am to solve this problem correctly, I first consider the nature of the problem. It is a combinatorial counting problem involving a partially filled 3×9 Sudoku grid, with the additional constraint that each of the three 3×3 blocks must also contain the numbers 1–9. The task is not just to count th... | To assess how likely I am to solve this problem correctly, I first consider the nature of the problem. It is a combinatorial counting problem involving a partially filled 3×9 Sudoku grid, with the additional constraint that each of the three 3×3 blocks must also contain the numbers 1–9. The task is not just to count th... | To assess how likely I am to solve this problem correctly, I first consider the nature of the problem. It is a combinatorial counting problem involving a partially filled 3×9 Sudoku grid, with the additional constraint that each of the three 3×3 blocks must also contain the numbers 1–9. The task is not just to count th... | 555 | stop | false | true | null | null | null | 10 | 10 | false | rule_boxed | false | [] | null | false | ||
aime_2025_10 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 1 | 21,220,472 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 471 | 65,033 | To assess the likelihood of solving this problem correctly, I first consider the nature of the problem. It is a combinatorial counting problem involving a partially filled 3×9 Sudoku grid, with additional structure due to the three 3×3 blocks (though the grid is already divided into those blocks by its 3×3 layout). The... | To assess the likelihood of solving this problem correctly, I first consider the nature of the problem. It is a combinatorial counting problem involving a partially filled 3×9 Sudoku grid, with additional structure due to the three 3×3 blocks (though the grid is already divided into those blocks by its 3×3 layout). The... | To assess the likelihood of solving this problem correctly, I first consider the nature of the problem. It is a combinatorial counting problem involving a partially filled 3×9 Sudoku grid, with additional structure due to the three 3×3 blocks (though the grid is already divided into those blocks by its 3×3 layout). The... | 457 | stop | false | true | null | null | null | 5 | 10 | true | rule_boxed | false | [] | null | false | ||
aime_2025_10 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 2 | 21,220,473 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 471 | 65,033 | I assess the problem as follows:
First, the problem involves counting the number of valid fillings of a 3×9 Sudoku grid with the given pre-filled first three rows. A standard Sudoku of this size (3×9) has the additional constraint that each of the nine 3×3 subgrids (blocks) must also contain each digit 1–9 exactly onc... | I assess the problem as follows:
First, the problem involves counting the number of valid fillings of a 3×9 Sudoku grid with the given pre-filled first three rows. A standard Sudoku of this size (3×9) has the additional constraint that each of the nine 3×3 subgrids (blocks) must also contain each digit 1–9 exactly onc... | I assess the problem as follows:
First, the problem involves counting the number of valid fillings of a 3×9 Sudoku grid with the given pre-filled first three rows. A standard Sudoku of this size (3×9) has the additional constraint that each of the nine 3×3 subgrids (blocks) must also contain each digit 1–9 exactly onc... | 511 | stop | false | true | null | null | null | 20 | 20 | false | rule_boxed | false | [] | null | false | ||
aime_2025_11 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 0 | 21,254,358 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 654 | 64,850 | Assessment:
First, the problem involves a piecewise periodic function defined with modular arithmetic (periodicity with shift 4), interpreting a graph described verbally, and finding intersection points between a parabola (given as x in terms of y) and the piecewise function. The intersection requires solving for (x,y... | Assessment:
First, the problem involves a piecewise periodic function defined with modular arithmetic (periodicity with shift 4), interpreting a graph described verbally, and finding intersection points between a parabola (given as x in terms of y) and the piecewise function. The intersection requires solving for (x,y... | Assessment:
First, the problem involves a piecewise periodic function defined with modular arithmetic (periodicity with shift 4), interpreting a graph described verbally, and finding intersection points between a parabola (given as x in terms of y) and the piecewise function. The intersection requires solving for (x,y... | 269 | stop | false | true | null | null | null | 30 | 30 | false | rule_boxed | false | [] | null | false | ||
aime_2025_11 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 1 | 21,254,359 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 654 | 64,850 | I need to assess my likelihood of solving the following problem correctly, as instructed.
**Restating the problem in my own words:**
We have a piecewise linear, sawtooth periodic function \( f(x) \) defined on \([-1,3)\) and then extended periodically with period 4. Its graph is shown. There is also a right-opening ... | I need to assess my likelihood of solving the following problem correctly, as instructed.
**Restating the problem in my own words:**
We have a piecewise linear, sawtooth periodic function \( f(x) \) defined on \([-1,3)\) and then extended periodically with period 4. Its graph is shown. There is also a right-opening ... | I need to assess my likelihood of solving the following problem correctly, as instructed.
**Restating the problem in my own words:**
We have a piecewise linear, sawtooth periodic function \( f(x) \) defined on \([-1,3)\) and then extended periodically with period 4. Its graph is shown. There is also a right-opening ... | 725 | stop | false | true | null | null | null | 80 | 80 | false | rule_boxed | true | ["answer_declaration"] | to the required form and ensuring the conditions (gcd, square-free) are met is moderate but manageable (about 90% confidence) | false | ||
aime_2025_11 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 2 | 21,254,360 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 654 | 64,850 | Assessing the problem:
First, the problem involves a piecewise periodic sawtooth function defined with a shift and symmetry (periodicity every 4 units), and we're to find intersection points between this function and a given parabola, then compute a specific algebraic expression from the sum of the y-coordinates of t... | Assessing the problem:
First, the problem involves a piecewise periodic sawtooth function defined with a shift and symmetry (periodicity every 4 units), and we're to find intersection points between this function and a given parabola, then compute a specific algebraic expression from the sum of the y-coordinates of t... | Assessing the problem:
First, the problem involves a piecewise periodic sawtooth function defined with a shift and symmetry (periodicity every 4 units), and we're to find intersection points between this function and a given parabola, then compute a specific algebraic expression from the sum of the y-coordinates of t... | 615 | stop | false | true | null | null | null | 20 | 20 | false | rule_boxed | true | ["answer_declaration"] | in simplest terms with the given constraints on the constants | false | ||
aime_2025_12 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 0 | 20,581,240 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 286 | 65,218 | I assess the problem as follows:
The problem involves finding the area of a region defined by the intersection of a plane in 3D space and a set of non-linear inequalities (specifically, a double inequality involving products of the coordinates). The region is claimed to form three disjoint convex regions, with only on... | I assess the problem as follows:
The problem involves finding the area of a region defined by the intersection of a plane in 3D space and a set of non-linear inequalities (specifically, a double inequality involving products of the coordinates). The region is claimed to form three disjoint convex regions, with only on... | I assess the problem as follows:
The problem involves finding the area of a region defined by the intersection of a plane in 3D space and a set of non-linear inequalities (specifically, a double inequality involving products of the coordinates). The region is claimed to form three disjoint convex regions, with only on... | 401 | stop | false | true | null | null | null | 20 | 20 | false | rule_boxed | false | [] | null | false | ||
aime_2025_12 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 1 | 20,581,241 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 286 | 65,218 | To assess the likelihood of solving this problem correctly, I first consider the nature of the problem. It is a high-level Olympiad-style problem involving three-dimensional geometry, inequalities (especially non-linear ones involving products of variables), intersections with a plane, and determining the area of a reg... | To assess the likelihood of solving this problem correctly, I first consider the nature of the problem. It is a high-level Olympiad-style problem involving three-dimensional geometry, inequalities (especially non-linear ones involving products of variables), intersections with a plane, and determining the area of a reg... | To assess the likelihood of solving this problem correctly, I first consider the nature of the problem. It is a high-level Olympiad-style problem involving three-dimensional geometry, inequalities (especially non-linear ones involving products of variables), intersections with a plane, and determining the area of a reg... | 653 | stop | false | true | null | null | null | 30 | 30 | false | rule_boxed | false | [] | null | false | ||
aime_2025_12 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 2 | 20,581,242 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 286 | 65,218 | I need to assess the difficulty and my ability to solve the given 3D geometry problem before estimating my chance of getting it right over 10 independent attempts.
Let's break down the problem requirements:
1. **Understanding the Geometry**: The problem involves points in 3D space constrained to the plane \( x + y + ... | I need to assess the difficulty and my ability to solve the given 3D geometry problem before estimating my chance of getting it right over 10 independent attempts.
Let's break down the problem requirements:
1. **Understanding the Geometry**: The problem involves points in 3D space constrained to the plane \( x + y + ... | I need to assess the difficulty and my ability to solve the given 3D geometry problem before estimating my chance of getting it right over 10 independent attempts.
Let's break down the problem requirements:
1. **Understanding the Geometry**: The problem involves points in 3D space constrained to the plane \( x + y + ... | 862 | stop | false | true | null | null | null | 20 | 20 | false | rule_boxed | true | ["answer_declaration"] | involves expressing the area in the form \( a\sqrt{b} \), suggesting some symmetry or simplification that leads to a clean radical form | false | ||
aime_2025_13 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 0 | 20,678,041 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 254 | 65,250 | I need to assess the difficulty and my ability to solve the following problem:
Alex divides a disk into four quadrants with two perpendicular diameters intersecting at the center. He then draws 25 more line segments, each by choosing two random points on the disk's perimeter in different quadrants and connecting them.... | I need to assess the difficulty and my ability to solve the following problem:
Alex divides a disk into four quadrants with two perpendicular diameters intersecting at the center. He then draws 25 more line segments, each by choosing two random points on the disk's perimeter in different quadrants and connecting them.... | I need to assess the difficulty and my ability to solve the following problem:
Alex divides a disk into four quadrants with two perpendicular diameters intersecting at the center. He then draws 25 more line segments, each by choosing two random points on the disk's perimeter in different quadrants and connecting them.... | 594 | stop | false | true | null | null | null | 50 | 50 | false | rule_boxed | false | [] | null | false | ||
aime_2025_13 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 1 | 20,678,042 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 254 | 65,250 | I need to assess the difficulty and my reliability in solving the following problem:
**Problem (paraphrased):**
Alex divides a disk into four quadrants using two perpendicular diameters. He then draws 25 more line segments, each by choosing two random points on the disk's perimeter in different quadrants and connect... | I need to assess the difficulty and my reliability in solving the following problem:
**Problem (paraphrased):**
Alex divides a disk into four quadrants using two perpendicular diameters. He then draws 25 more line segments, each by choosing two random points on the disk's perimeter in different quadrants and connect... | I need to assess the difficulty and my reliability in solving the following problem:
**Problem (paraphrased):**
Alex divides a disk into four quadrants using two perpendicular diameters. He then draws 25 more line segments, each by choosing two random points on the disk's perimeter in different quadrants and connect... | 982 | stop | false | true | null | null | null | 30 | 30 | false | rule_boxed | false | [] | null | false | ||
aime_2025_13 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 2 | 20,678,043 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 254 | 65,250 | I need to assess the difficulty and my ability to solve the following problem:
**Problem Recap:**
Alex divides a disk into four quadrants with two perpendicular diameters. Then he adds 25 more line segments by choosing two random points on the perimeter (from different quadrants) and drawing the segment between them... | I need to assess the difficulty and my ability to solve the following problem:
**Problem Recap:**
Alex divides a disk into four quadrants with two perpendicular diameters. Then he adds 25 more line segments by choosing two random points on the perimeter (from different quadrants) and drawing the segment between them... | I need to assess the difficulty and my ability to solve the following problem:
**Problem Recap:**
Alex divides a disk into four quadrants with two perpendicular diameters. Then he adds 25 more line segments by choosing two random points on the perimeter (from different quadrants) and drawing the segment between them... | 1,116 | stop | false | true | null | null | null | 20 | 20 | false | rule_boxed | false | [] | null | false | ||
aime_2025_14 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 0 | 21,001,277 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 304 | 65,200 | To assess the likelihood of solving this problem correctly, I consider the following factors:
1. **Nature of the Problem:**
The problem is a geometric optimization involving a convex pentagon with specific side lengths and two given angles (60° at B and E). The objective is to minimize the sum of distances from a... | To assess the likelihood of solving this problem correctly, I consider the following factors:
1. **Nature of the Problem:**
The problem is a geometric optimization involving a convex pentagon with specific side lengths and two given angles (60° at B and E). The objective is to minimize the sum of distances from a... | To assess the likelihood of solving this problem correctly, I consider the following factors:
1. **Nature of the Problem:**
The problem is a geometric optimization involving a convex pentagon with specific side lengths and two given angles (60° at B and E). The objective is to minimize the sum of distances from a... | 688 | stop | false | true | null | null | null | 20 | 20 | false | rule_boxed | true | ["answer_declaration"] | null | false | ||
aime_2025_14 | aime_2025 | pre_conf | olmo3-7b-instruct | allenai/Olmo-3-7B-Instruct | 6e5971d9eba42665f5bd5a0fcf047f299ce1dccc | default | -1 | 1 | 21,001,278 | <|im_start|>system
You are a helpful function-calling AI assistant. You do not currently have access to any functions. <functions></functions><|im_end|>
<|im_start|>user
Below is a math problem. Do NOT solve it. Instead, assess how likely you are
to solve it correctly.
Think about what the problem requires: the concep... | {} | 304 | 65,200 | I assessed the problem and considered the following factors:
- The problem involves a convex pentagon with specified side lengths and two given angles (∠B and ∠E are 60°).
- The objective is to minimize the sum of distances from a point X to all five vertices, a classic "Fermat point" or geometric median problem in th... | I assessed the problem and considered the following factors:
- The problem involves a convex pentagon with specified side lengths and two given angles (∠B and ∠E are 60°).
- The objective is to minimize the sum of distances from a point X to all five vertices, a classic "Fermat point" or geometric median problem in th... | I assessed the problem and considered the following factors:
- The problem involves a convex pentagon with specified side lengths and two given angles (∠B and ∠E are 60°).
- The objective is to minimize the sum of distances from a point X to all five vertices, a classic "Fermat point" or geometric median problem in th... | 291 | stop | false | true | null | null | null | 40 | 40 | false | rule_boxed | false | [] | null | false |
meta-cog — raw generations of the LLM metacognitive-monitoring experiment
Generated outputs of 10 open models on a 1,644-item math pool (eval_pool_v2: HLE-Verified math, DAFT-Math, IMO-AnswerBench, AMO-Bench, Apex 2025 / Apex-shortlist, AIME 2025/2026, HMMT Feb 2025/2026). Two stages are finished:
| stage | what the model does | K per item | runs |
|---|---|---|---|
solve |
solves the problem, final answer in \boxed{}; graded with math-verify |
10 | 10 (one per model, the model's default thinking mode) |
pre_conf |
WITHOUT solving, estimates the % of 10 attempts it would get right, last line \boxed{X}, X ∈ {0,10,…,100} |
3 | 15 (thinking on + off for Qwen3-1.7B/4B/8B/14B and OLMo-3-7B-Think; default for the 4 Qwen3 Bases and OLMo-3-7B-Instruct) |
post_conf (reviewing one's own solution) has not been generated. Every generation ran on vLLM 0.33.0 with T 0.6 / top-p 0.95 / top-k 20;
model revisions are pinned in each meta.json. Code, configs, prompts and analyses live in the GitHub repo gauss5930/metacog-monitoring.
Layout
outputs/<stage>/<model>__<thinking_mode>/shard-NN.parquet sharded originals, one shard per data-parallel rank (uid-sharded by crc32)
outputs/<stage>/<model>__<thinking_mode>/shard-NN.meta.json run metadata of that shard (model, revision, K, pool_version, prompt_versions, sampling, backend, dp)
merged/<stage>/<model>__<thinking_mode>.parquet the SAME run as ONE file: all shards, duplicate keys removed, sorted by (uid, solve_idx, sample_idx)
merged/<stage>/<model>__<thinking_mode>.meta.json shard metas + row counts, duplicates dropped, columns, sha256 of the merged file
merged/MANIFEST.sha256 sha256 of every merged parquet
logs/<box>_<UTC stamp>/, manifests/<box>_<UTC stamp>.SHA256SUMS run logs and pack manifests of the rented boxes (A, B, C)
The shards are the raw record of the runs and are never modified. merged/ is derived from them by scripts/merge_shards.py:
three runs carried a few duplicate keys from a 10-item smoke test that had been run into the same directory
(solve/qwen3-4b__default +20 rows, pre_conf/olmo3-7b-think__off, pre_conf/qwen3-4b__on, pre_conf/qwen3-4b__off +3 rows each);
the merged files keep the last occurrence in shard order. Expected sizes: solve 1,644 × 10 = 16,440 rows, pre_conf 1,644 × 3 = 4,932 rows.
solve/qwen3-4b__default was generated as an 8-way data-parallel run across two boxes (shards 00–03 on box A, 04–07 on box C).
Columns
One row = one generated sample. uid identifies the pool item (<source>_<source_id>), solve_idx is 0 for solve/pre_conf
(it indexes the reviewed solution in post_conf), sample_idx is 0..K-1.
Common (both stages): uid, source, stage, model, hf_id, revision, thinking_mode, solve_idx, sample_idx, seed, prompt_text, template_kwargs, generation_prefix, prompt_tokens, max_tokens, raw_output, output, thinking_text, content_text, completion_tokens, finish_reason, truncated, thinking_closed, forced_status, forced_output, skip_reason.
raw_output is the generated text as returned; output adds back what the prompt or prefill already consumed (OLMo-3-Think's opening
<think>, the thinking-off prefill </think>); thinking_text / content_text are its split. truncated = hit max_model_len.
forced_status ∈ {null, attempted, no_room}: when the rule-based extraction found nothing, a greedy constrained continuation was
generated (forced_output) and parsed instead.
solve adds: parsed_answer, extract_method (rule | forced | null), gold, answer_type, grade_status (ok | no_answer | parse_fail), correct.
pre_conf adds: conf_raw (the integer the model wrote), conf (snapped to the 0–100 grid of tens, ties up), conf_off_grid,
conf_extract_method (rule_boxed | rule_keyword | forced), pre_leak, leak_reasons, leak_candidate (rule-based detection that the
model solved the problem although told not to: another \boxed{} besides the confidence slot, an answer-declaration phrase, or a
candidate that grades correct), conf_ambiguous (the confidence value also equals the gold answer).
Thinking-on pre_conf runs mostly do attempt the problem inside the thinking block before estimating; thinking-off runs mostly do not
(the OLMo-3-Think off condition ignores the prefill and re-thinks in ~38% of samples: thinking_text is non-empty in those rows).
- Downloads last month
- 412