Dataset Viewer
Auto-converted to Parquet Duplicate
challenge_id
stringclasses
487 values
source
stringclasses
1 value
split
stringclasses
2 values
theorem_name
stringclasses
487 values
challenge_lean
stringclasses
487 values
prover_id
stringclasses
1 value
attempt
int64
0
39
stop_reason
stringclasses
1 value
trace
stringlengths
1.17k
47.5k
grade
stringclasses
2 values
grade_detail
stringclasses
3 values
curation_status
stringclasses
4 values
n_valid
int64
0
37
n_invalid
int64
0
39
interval_low
float64
0
0.84
interval_high
float64
0.08
1
aimeII_2001_p3
minif2f
validation
aimeII_2001_p3
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
0
eos
### Detailed Proof and Analysis First, we need to understand the problem. We have a sequence of integers `x(n)` defined for all natural numbers `n`, with initial values: - `x(1) = 211`, - `x(2) = 375`, - `x(3) = 420`, - `x(4) = 523`, and for `n ≥ 5`, it satisfies the recurrence relation: `x(n) = x(n - 1) - x(n - 2) + ...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeII_2001_p3
minif2f
validation
aimeII_2001_p3
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
1
eos
### Detailed Proof and Analysis First, we need to determine the values of `x 5`, `x 6`, ..., `x 975` using the recurrence relation: \[ x_n = x_{n-1} - x_{n-2} + x_{n-3} - x_{n-4} \] for \( n \geq 5 \). The initial conditions are: 1. \( x_1 = 211 \) 2. \( x_2 = 375 \) 3. \( x_3 = 420 \) 4. \( x_4 = 523 \) #### Step 1:...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeII_2001_p3
minif2f
validation
aimeII_2001_p3
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
2
eos
### Detailed Proof and Analysis First, we observe that the problem involves a recurrence relation for a sequence of integers indexed by natural numbers. The recurrence is not straightforward, as it's non-linear and depends on the previous four terms. The initial terms are given, and we need to compute \(x_{531} + x_{7...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeII_2020_p6
minif2f
validation
aimeII_2020_p6
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
0
eos
### Detailed Proof and Analysis First, we need to understand the problem. We have a sequence \( t_n \) defined by: - \( t_1 = 20 \) - \( t_2 = 21 \) - For \( n \geq 3 \), \( t_n = \frac{5 \cdot t_{n-1} + 1}{25 \cdot t_{n-2}} \) We are to compute \( \text{den}(t_{2020}) + \text{num}(t_{2020}) \) where: 1. \( \text{den...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeII_2020_p6
minif2f
validation
aimeII_2020_p6
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
1
eos
### Detailed Proof and Analysis First, let's understand the problem and the sequence `t(n)`: 1. **Initial Terms**: - `t(1) = 20` - `t(2) = 21` 2. **Recurrence Relation for `n ≥ 3`**: - `t(n) = (5 * t(n - 1) + 1) / (25 * t(n - 2))` Our goal is to compute `t(2020)` and show that its numerator and denominator...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeII_2020_p6
minif2f
validation
aimeII_2020_p6
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
2
eos
### Detailed Proof and Analysis First, we observe the sequence defined by: - \( t_1 = 20 \), - \( t_2 = 21 \), - For \( n \geq 3 \), \( t_n = \frac{5 \cdot t_{n-1} + 1}{25 \cdot t_{n-2}} \). We need to compute the denominator and numerator of \( t_{2020} \) (in reduced form) and show that \( \text{den} + \text{num} =...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeI_2000_p7
minif2f
validation
aimeI_2000_p7
import Mathlib.Algebra.BigOperators.Basic import Mathlib.Data.Real.Basic import Mathlib.Data.Complex.Basic import Mathlib.Data.Nat.Log import Mathlib.Data.Complex.Exponential import Mathlib.NumberTheory.Divisors import Mathlib.Data.ZMod.Defs import Mathlib.Data.ZMod.Basic import Mathlib.Topology.Basic import Mathlib.Da...
deepseek-ai/DeepSeek-Prover-V2-7B
0
eos
### Detailed Proof and Analysis First, we need to understand the problem and the given conditions. We have three positive real numbers \( x, y, z \) such that: 1. \( x \cdot y \cdot z = 1 \), 2. \( x + \frac{1}{z} = 5 \), 3. \( y + \frac{1}{x} = 29 \), 4. \( z + \frac{1}{y} = m \) (where \( m \) is a rational number w...
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeI_2000_p7
minif2f
validation
aimeI_2000_p7
"import Mathlib.Algebra.BigOperators.Basic\nimport Mathlib.Data.Real.Basic\nimport Mathlib.Data.Comp(...TRUNCATED)
deepseek-ai/DeepSeek-Prover-V2-7B
1
eos
"### Detailed Proof and Analysis\n\nFirst, let's carefully restate and understand the problem.\n\n**(...TRUNCATED)
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aimeI_2000_p7
minif2f
validation
aimeI_2000_p7
"import Mathlib.Algebra.BigOperators.Basic\nimport Mathlib.Data.Real.Basic\nimport Mathlib.Data.Comp(...TRUNCATED)
deepseek-ai/DeepSeek-Prover-V2-7B
2
eos
"### Detailed Proof and Analysis\n\n#### Understanding the Problem\nWe have three positive real numb(...TRUNCATED)
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
aime_1983_p1
minif2f
test
aime_1983_p1
"import Mathlib.Algebra.BigOperators.Basic\nimport Mathlib.Data.Real.Basic\nimport Mathlib.Data.Comp(...TRUNCATED)
deepseek-ai/DeepSeek-Prover-V2-7B
0
eos
"### Detailed Proof and Analysis\n\nFirst, we need to understand the problem and the given condition(...TRUNCATED)
invalid
lean did not elaborate cleanly (status=error)
single_class
0
3
0
0.195172
End of preview. Expand in Data Studio

MiniF2F DeepSeek-Prover-V2 Traces

One row per proof attempt (trace) produced while screening MiniF2F problems with DeepSeek-Prover-V2-7B. Each row carries the generated proof text, its Lean grade, and the per-challenge curation decision.

Provenance

MiniF2F theorem statements -> DeepSeek-Prover-V2-7B generations (multiple seeded attempts per problem) -> Lean elaboration + comparator grading (valid / invalid / incomplete) -> per-challenge curation from a Beta(0.1 + n_valid, 0.1 + n_invalid) posterior on the validity rate, whose credible interval decides each challenge's status.

Columns

column description
challenge_id MiniF2F problem id the trace attempts.
source Constant "minif2f"; the upstream statement source.
split MiniF2F split the problem belongs to (validation or test).
theorem_name Name of the theorem/lemma the prover must complete.
challenge_lean Self-contained Lean file (Mathlib header + theorem ... := by sorry) given to the prover.
prover_id Constant "deepseek-ai/DeepSeek-Prover-V2-7B"; the model that generated the trace.
attempt Zero-based attempt index for this challenge (each attempt a fresh seed).
stop_reason Why generation stopped (e.g. stop, max_new_tokens).
trace Full generated text (the model's proof attempt).
grade Three-way grade from Lean elaboration: valid, invalid, or incomplete.
grade_detail Human-readable grader detail for the grade.
curation_status Per-challenge curation outcome: accepted, needs_more_info, out_of_band, or single_class.
n_valid Per-challenge count of valid grades feeding the curation posterior.
n_invalid Per-challenge count of invalid grades feeding the curation posterior.
interval_low Low bound of the per-challenge Beta-posterior credible interval on the validity rate.
interval_high High bound of the per-challenge Beta-posterior credible interval on the validity rate.

Distributions

  • Total rows (traces): 3940
  • Distinct challenges: 487

Rows per grade:

  • invalid: 2404
  • valid: 1536

Rows per split:

  • test: 2135
  • validation: 1805

Curation status over challenges:

  • accepted: 24
  • needs_more_info: 27
  • out_of_band: 16
  • single_class: 420

Curation status over rows:

  • accepted: 960
  • needs_more_info: 1080
  • out_of_band: 640
  • single_class: 1260

Generation config

Trace-generation configuration used to produce this dataset:

  • prover_id: deepseek-ai/DeepSeek-Prover-V2-7B
  • max_new_tokens: 32768
  • temperature: 1.0
  • top_p: 0.95
  • base_seed: 0
  • seed_per_attempt: base_seed + attempt
  • max_model_len: 16384
  • dtype: auto

Caveats

  • incomplete grades are token-limited generations that produced no gradeable proof; they form no valid/invalid evidence and are excluded from the curation posterior counts.
  • single_class challenges never produced a contrasting pair (one of n_valid or n_invalid is zero), so no within-challenge validity contrast exists.
  • out_of_band challenges do have a contrasting pair, but their validity-rate credible interval fell outside the informative band [0.2, 0.8].
  • split is recomputed by loading MiniF2F; a challenge absent from both splits is left null.
Downloads last month
45