Datasets:

GlimmaryKarl
/

DualBlind

id stringlengths 21 29	problem_id stringclasses 36 values	problem_title stringclasses 35 values	suite stringclasses 11 values	topic stringclasses 7 values	domain stringclasses 35 values	difficulty stringclasses 4 values	model stringclasses 386 values	system stringclasses 1 value	prompt stringclasses 36 values	messages listlengths 3 3	canonical_answer stringclasses 35 values	is_verified bool 1 class	accuracy_score int64 100 100	turns_to_consensus int64 1 24	efficiency_index float64 0 536	cost_usd float64 0 0.24
sft-cli-run-1788794795775	gpqa-diamond-01	3D Isotropic Quantum Harmonic Oscillator Degeneracy	GPQA Diamond	science	Quantum Mechanics & Degeneracy	Extreme	meta-llama/llama-3.3-70b-instruct:free + qwen/qwq-32b:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Consider a particle of mass $m$ trapped in a three-dimensional isotropic quantum harmonic oscillator potential $V(r) = \frac{1}{2} m \omega^2 (x^2 + y^2 + z^2)$. The total energy eigenvalues are quantized according to: $$E_n = \hbar \omega \left(n + \frac{3}{2}\right), \quad n = n_x + n_y + n_z \in \{0, 1, 2, 3, \dots\...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Consider a particle of mass $m...	21 states	true	100	2	16.75	0.001147
sft-cli-run-1788794774532	mmlu-pro-02	Byzantine Agreement Quorum Intersection Invariant	MMLU-Pro	science	Distributed Systems & Byzantine Fault Tolerance	Medium	openrouter/free + gemini-1.5-flash	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In a distributed consensus network running the classical PBFT (Practical Byzantine Fault Tolerance) protocol with $N$ total replica nodes, up to $f$ nodes may be Byzantine (arbitrarily malicious, colluding, or crashing). For the protocol to maintain safety (no two non-faulty nodes commit conflicting values) and livenes...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In a distributed consensus net...	13 nodes	true	100	2	11.96	0.001105
sft-cli-run-1788794746174	mmlu-pro-01	Relativistic Carnot Engine Efficiency with Heat Exchange	MMLU-Pro	science	Physics & Thermodynamics	Hard	qwen/qwen-2.5-coder-32b-instruct:free + gemini-2.5-flash	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A hypothetical relativistic heat engine operates between two thermal reservoirs with temperatures $T_H = 1200\text{ K}$ and $T_C = 300\text{ K}$. During each cycle, the working fluid absorbs $Q_H = 4800\text{ J}$ of heat from the hot reservoir. Due to finite-time heat transfer kinetics across the boundary thermal condu...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A hypothetical relativistic he...	Efficiency = 50%, Work = 2400 J	true	100	2	28.32	0.000784
sft-cli-run-1788751473740	frontiermath-01	Congruent Number Elliptic Curve Mordell-Weil Torsion & 2-Selmer Rank	FrontierMath	math	Arithmetic Geometry & Elliptic Curves	Extreme	qwen/qwen-2.5-72b-instruct:free + qwen/qwq-32b:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Consider the family of congruent number elliptic curves defined over $\mathbb{Q}$ by: $$E_d: y^2 = x^3 - d^2 x$$ where $d$ is a square-free positive integer. By the Mordell-Weil theorem, the group of rational points is finitely generated: $E_d(\mathbb{Q}) \cong \mathbb{Z}^r \oplus E_d(\mathbb{Q})_{\text{tors}}$. 1. Wha...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Consider the family of congrue...	Torsion Group: Z/2Z x Z/2Z (order 4), Rank r: 1	true	100	2	2.55	0.002336
sft-cli-run-1788749591784	gpqa-diamond-05	Crick Wobble Hypothesis Inosine Base Pairing & tRNA Synthetase Universality	GPQA Diamond	science	Molecular Genetics & Translation Wobble	Hard	deepseek/deepseek-r1:free + meta-llama/llama-3.3-70b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In standard biological translation across living organisms: 1. In Francis Crick's classic Wobble Hypothesis, post-transcriptional deamination of adenosine at the first (5') anticodon position produces the non-canonical modified nucleoside Inosine (I). What are ALL three mRNA 3'-codon bases that Inosine in the wobble po...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In standard biological transla...	Wobble Bases: A, U, C, Synthetases: 20	true	100	3	4.58	0.001542
sft-cli-run-1788749457243	gpqa-diamond-02	Competitive Enzyme Inhibition Apparent Km Ratio	GPQA Diamond	science	Biochemistry & Michaelis-Menten Kinetics	Hard	meta-llama/llama-3.3-70b-instruct:free + deepseek/deepseek-r1:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	An enzyme follows standard Michaelis-Menten kinetics with substrate $S$, having an uninhibited Michaelis constant $K_m = 4.0\text{ \mu M}$ and $V_{\text{max}} = 100\text{ \mu M/s}$. A competitive inhibitor $I$ with dissociation constant $K_i = 2.0\text{ \mu M}$ is added at a concentration $[I] = 6.0\text{ \mu M}$. In ...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "An enzyme follows standard Mic...	Km_app = 16 μM, v = 50 μM/s	true	100	3	6	0.001556
sft-cli-run-1788749129492	abstract-04	The Monty Hall Variant with 4 Doors and 2 Car Prizes	Game Theory	strategy	Bayesian Probability & Counterfactual Switching	Hard	gemini-1.5-flash + qwen/qwen-2.5-72b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A game show has 4 closed doors. Behind 2 of the doors are brand-new Luxury Cars, and behind the other 2 doors are Goats. All distributions are equally likely. 1. The contestant picks Door #1. 2. The host, who knows what is behind every door, opens ONE of the remaining doors (Door #2, #3, or #4) that contains a Goat. 3...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A game show has 4 closed doors...	3/4	true	100	2	10.62	0.001088
sft-cli-run-1788748803413	strategy-02	Cournot Duopoly Equilibrium Pricing	Game Theory	strategy	Oligopoly Pricing & Nash Equilibrium	Hard	microsoft/phi-3-mini-128k-instruct:free + qwen/qwen-2.5-coder-32b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Two identical competing software firms, Firm A and Firm B, simultaneously choose their output quantities $q_A$ and $q_B$ of an AI API service. The inverse market demand function is given by: $P = 120 - (q_A + q_B)$ Both firms have an identical constant marginal cost of $MC = 12$ per unit and zero fixed costs. Each fir...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Two identical competing softwa...	qA = 36, qB = 36, Price = $48	true	100	2	12.7	0.001128
sft-cli-run-1788748767796	arc-challenge-01	3x3 Matrix 90-Degree Orthogonal Rotation & Bitwise XOR Merge	ARC Challenge	abstract	Spatial Invariance & Inductive Logic	Medium	deepseek/deepseek-chat:free + openrouter/free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Given a binary 3x3 input matrix $M$: Row 1: [1, 0, 0] Row 2: [1, 1, 0] Row 3: [0, 0, 1] We perform two consecutive deterministic transformations: 1. Step 1: Rotate matrix $M$ clockwise by $90^\circ$ to produce $M_{\text{rot}}$. 2. Step 2: Compute the element-wise XOR matrix $R = M \oplus M_{\text{rot}}$, where $a \opl...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Given a binary 3x3 input matri...	6 active bits	true	100	2	9.64	0.00146
sft-cli-run-1788748630004	swe-bench-05	scikit-learn Ridge Regression: Gram Matrix Condition Number Squaring	SWE-bench	coding	Numerical Stability & Machine Learning Estimators	Medium	microsoft/phi-3-mini-128k-instruct:free + qwen/qwq-32b:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In scikit-learn issue scikit-learn__scikit-learn-14201, resolving numerical precision regressions in `Ridge` regression: Given an ill-conditioned design matrix $X \in \mathbb{R}^{n \times p}$ with condition number $\kappa_2(X) = 10^5$: 1. If an estimator solves the normal equations directly via Cholesky decomposition o...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In scikit-learn issue scikit-l...	Gram Condition Number: 10^10, LAPACK Driver: gelsd	true	100	2	4.4	0.001849
sft-cli-run-1788748573320	swe-bench-04	SymPy Defective Matrix Jordan Block Decomposition & Minimal Polynomial	SWE-bench	coding	Computer Algebra Systems & Symbolic Linear Algebra	Hard	qwen/qwen-2.5-coder-32b-instruct:free + google/gemini-2.0-flash-exp:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In SymPy issue sympy__sympy-19453, evaluating `Matrix.jordan_form()` on defective nilpotent and degenerate matrices triggered algebraic infinite recursion. Consider the $3 \times 3$ rational matrix: $$A = \begin{pmatrix} 3 & 1 & 0 \\ 0 & 3 & 1 \\ 0 & 0 & 3 \end{pmatrix}$$ 1. What is the geometric multiplicity $\dim(\ke...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In SymPy issue sympy__sympy-19...	Geometric Multiplicity: 1, Jordan Block Size: 3x3, Minimal Polynomial: (lambda - 3)^3, (A - 3I)^3: 0 (Zero Matrix)	true	100	2	6.01	0.001579
sft-cli-run-1788748507703	gpqa-diamond-05	Crick Wobble Hypothesis Inosine Base Pairing & tRNA Synthetase Universality	GPQA Diamond	science	Molecular Genetics & Translation Wobble	Hard	deepseek/deepseek-r1-distill-llama-70b:free + meta-llama/llama-3.1-8b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In standard biological translation across living organisms: 1. In Francis Crick's classic Wobble Hypothesis, post-transcriptional deamination of adenosine at the first (5') anticodon position produces the non-canonical modified nucleoside Inosine (I). What are ALL three mRNA 3'-codon bases that Inosine in the wobble po...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In standard biological transla...	Wobble Bases: A, U, C, Synthetases: 20	true	100	2	6.02	0.001455
sft-cli-run-1788747725681	strategy-01	The 21-Token Subtraction Game (Nim Variant)	Game Theory	strategy	Combinatorial Nim Variants	Easy	google/gemini-2.0-flash-exp:free + meta-llama/llama-3.1-8b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Two players play a turn-based mathematical game with a pile of 21 tokens. - On each turn, a player MUST remove either 1, 2, or 3 tokens from the pile. - The player who takes the LAST remaining token wins the entire game. - Player 1 goes first. Both players have perfect information and play with flawless game-theoretic ...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Two players play a turn-based ...	1 token	true	100	2	5.45	0.001506
sft-cli-run-1788747632191	arc-challenge-02	2D Cellular Lattice Transformation Rule	ARC Challenge	abstract	Inductive Pattern Topology	Hard	mistralai/mistral-7b-instruct:free + deepseek/deepseek-r1:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Consider a 3x3 binary grid state that evolves according to a deterministic local rule: Rule: At step $t+1$, cell $(r,c)$ becomes 1 if and only if the sum of its orthogonal Manhattan neighbors (Up, Down, Left, Right) at step $t$ is strictly ODD (1 or 3). Otherwise it becomes 0. Outside borders are strictly 0. Initial s...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Consider a 3x3 binary grid sta...	0 active cells	true	100	2	3.24	0.002009
sft-cli-run-1788747553284	ifeval-02	Verifiable 20-Word Constrained Summary with Numerical Invariants	IFEval	instruction_following	Exact Word-Count & Structural Invariants	Hard	mistralai/mistral-7b-instruct:free + deepseek/deepseek-chat:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Construct a summary of the dual-agent consensus mechanism that satisfies EVERY one of these 3 rules: 1. The summary text must contain EXACTLY 20 words (counted by whitespace separation). 2. The first word MUST be "Two" and the twentieth (last) word MUST be "truth". 3. The summary MUST mention the exact number "100%" wi...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Construct a summary of the dua...	Two autonomous neural agents communicate iteratively through multi-turn dialectic consensus until 100% agreement verifies the rigorous mathematical proof of truth	true	100	2	19.03	0.000805
sft-cli-run-1788747453580	swe-bench-05	scikit-learn Ridge Regression: Gram Matrix Condition Number Squaring	SWE-bench	coding	Numerical Stability & Machine Learning Estimators	Medium	qwen/qwq-32b:free + mistralai/mistral-7b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In scikit-learn issue scikit-learn__scikit-learn-14201, resolving numerical precision regressions in `Ridge` regression: Given an ill-conditioned design matrix $X \in \mathbb{R}^{n \times p}$ with condition number $\kappa_2(X) = 10^5$: 1. If an estimator solves the normal equations directly via Cholesky decomposition o...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In scikit-learn issue scikit-l...	Gram Condition Number: 10^10, LAPACK Driver: gelsd	true	100	2	3.7	0.001767
sft-cli-run-1788746330587	arc-challenge-01	3x3 Matrix 90-Degree Orthogonal Rotation & Bitwise XOR Merge	ARC Challenge	abstract	Spatial Invariance & Inductive Logic	Medium	microsoft/phi-3-mini-128k-instruct:free + deepseek/deepseek-chat:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Given a binary 3x3 input matrix $M$: Row 1: [1, 0, 0] Row 2: [1, 1, 0] Row 3: [0, 0, 1] We perform two consecutive deterministic transformations: 1. Step 1: Rotate matrix $M$ clockwise by $90^\circ$ to produce $M_{\text{rot}}$. 2. Step 2: Compute the element-wise XOR matrix $R = M \oplus M_{\text{rot}}$, where $a \opl...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Given a binary 3x3 input matri...	6 active bits	true	100	2	4.46	0.001686
sft-cli-run-1788746120405	swe-bench-03	Django ORM QuerySet: values() and annotate() GROUP BY Duplication Bug	SWE-bench	coding	ORM Query Compilation & SQL Invariants	Hard	microsoft/phi-3-mini-128k-instruct:free + openrouter/free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In a production Django repository (tested under SWE-bench issue django__django-13449 / #31824), an engineer writes the following query to count unique publications per author: ```python Author.objects.values('id').annotate(book_count=Count('books__id', distinct=True)).order_by('pub_date') ``` 1. Because the `Meta.order...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In a production Django reposit...	Bug: Appends pub_date to GROUP BY clause, Method: order_by()	true	100	2	10.16	0.000966
sft-cli-run-1788745774804	mmlu-pro-01	Relativistic Carnot Engine Efficiency with Heat Exchange	MMLU-Pro	science	Physics & Thermodynamics	Hard	openrouter/free + deepseek/deepseek-r1:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A hypothetical relativistic heat engine operates between two thermal reservoirs with temperatures $T_H = 1200\text{ K}$ and $T_C = 300\text{ K}$. During each cycle, the working fluid absorbs $Q_H = 4800\text{ J}$ of heat from the hot reservoir. Due to finite-time heat transfer kinetics across the boundary thermal condu...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A hypothetical relativistic he...	Efficiency = 50%, Work = 2400 J	true	100	2	10.13	0.001053
sft-cli-run-1788745499419	hle-01	CRISPR-Cas9 NHEJ Indel Out-of-Frame Frameshift Probability	Humanity's Last Exam	science	Molecular Biology & CRISPR Repair Dynamics	Extreme	meta-llama/llama-3.3-70b-instruct:free + deepseek/deepseek-chat:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In targeted genome editing, SpCas9 generates a double-strand break (DSB) 3 base pairs upstream of the NGG PAM sequence within a protein-coding exon. Repair proceeds predominantly via Non-Homologous End Joining (NHEJ) resulting in small insertions or deletions (indels). Assuming an idealized uniform prior distribution a...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In targeted genome editing, Sp...	Frameshift Probability = 2/3, MMEJ Modulo Class = 2	true	100	3	5.93	0.001314
sft-cli-run-1788745134945	strategy-02	Cournot Duopoly Equilibrium Pricing	Game Theory	strategy	Oligopoly Pricing & Nash Equilibrium	Hard	qwen/qwq-32b:free + deepseek/deepseek-chat:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Two identical competing software firms, Firm A and Firm B, simultaneously choose their output quantities $q_A$ and $q_B$ of an AI API service. The inverse market demand function is given by: $P = 120 - (q_A + q_B)$ Both firms have an identical constant marginal cost of $MC = 12$ per unit and zero fixed costs. Each fir...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Two identical competing softwa...	qA = 36, qB = 36, Price = $48	true	100	2	13.53	0.00085
sft-cli-run-1788745099726	arc-challenge-01	3x3 Matrix 90-Degree Orthogonal Rotation & Bitwise XOR Merge	ARC Challenge	abstract	Spatial Invariance & Inductive Logic	Medium	qwen/qwq-32b:free + openrouter/free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Given a binary 3x3 input matrix $M$: Row 1: [1, 0, 0] Row 2: [1, 1, 0] Row 3: [0, 0, 1] We perform two consecutive deterministic transformations: 1. Step 1: Rotate matrix $M$ clockwise by $90^\circ$ to produce $M_{\text{rot}}$. 2. Step 2: Compute the element-wise XOR matrix $R = M \oplus M_{\text{rot}}$, where $a \opl...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Given a binary 3x3 input matri...	6 active bits	true	100	2	8.74	0.001175
sft-cli-run-1788745062986	ifeval-02	Verifiable 20-Word Constrained Summary with Numerical Invariants	IFEval	instruction_following	Exact Word-Count & Structural Invariants	Hard	deepseek/deepseek-chat:free + qwen/qwen-2.5-coder-32b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Construct a summary of the dual-agent consensus mechanism that satisfies EVERY one of these 3 rules: 1. The summary text must contain EXACTLY 20 words (counted by whitespace separation). 2. The first word MUST be "Two" and the twentieth (last) word MUST be "truth". 3. The summary MUST mention the exact number "100%" wi...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Construct a summary of the dua...	Two autonomous neural agents communicate iteratively through multi-turn dialectic consensus until 100% agreement verifies the rigorous mathematical proof of truth	true	100	3	12.33	0.000943
sft-cli-run-1788744488682	logic-04	Strict Five-Person Seating Deduction	Formal Logic	logic	Linear Constraint Satisfaction	Hard	qwen/qwq-32b:free + qwen/qwen-2.5-72b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Five colleagues—Evelyn, Frank, Grace, Henry, and Ivy—are seated in a single row of 5 chairs numbered 1 to 5 from left to right. 1. Frank is sitting to the immediate left of Henry. 2. Evelyn is not sitting in chair 1 or chair 5. 3. Grace is sitting immediately adjacent to both Evelyn and Frank. 4. Ivy is sitting in an o...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Five colleagues—Evelyn, Frank,...	1:Ivy, 2:Evelyn, 3:Grace, 4:Frank, 5:Henry	true	100	3	4.45	0.001644
sft-cli-run-1788744341442	strategy-01	The 21-Token Subtraction Game (Nim Variant)	Game Theory	strategy	Combinatorial Nim Variants	Easy	qwen/qwen-2.5-coder-32b-instruct:free + qwen/qwq-32b:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Two players play a turn-based mathematical game with a pile of 21 tokens. - On each turn, a player MUST remove either 1, 2, or 3 tokens from the pile. - The player who takes the LAST remaining token wins the entire game. - Player 1 goes first. Both players have perfect information and play with flawless game-theoretic ...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Two players play a turn-based ...	1 token	true	100	2	9.78	0.001069
sft-cli-run-1788744230020	arc-challenge-01	3x3 Matrix 90-Degree Orthogonal Rotation & Bitwise XOR Merge	ARC Challenge	abstract	Spatial Invariance & Inductive Logic	Medium	qwen/qwen-2.5-72b-instruct:free + meta-llama/llama-3.3-70b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Given a binary 3x3 input matrix $M$: Row 1: [1, 0, 0] Row 2: [1, 1, 0] Row 3: [0, 0, 1] We perform two consecutive deterministic transformations: 1. Step 1: Rotate matrix $M$ clockwise by $90^\circ$ to produce $M_{\text{rot}}$. 2. Step 2: Compute the element-wise XOR matrix $R = M \oplus M_{\text{rot}}$, where $a \opl...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Given a binary 3x3 input matri...	6 active bits	true	100	2	8.8	0.001157
sft-cli-run-1788744192416	ifeval-02	Verifiable 20-Word Constrained Summary with Numerical Invariants	IFEval	instruction_following	Exact Word-Count & Structural Invariants	Hard	meta-llama/llama-3.3-70b-instruct:free + deepseek/deepseek-r1:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Construct a summary of the dual-agent consensus mechanism that satisfies EVERY one of these 3 rules: 1. The summary text must contain EXACTLY 20 words (counted by whitespace separation). 2. The first word MUST be "Two" and the twentieth (last) word MUST be "truth". 3. The summary MUST mention the exact number "100%" wi...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Construct a summary of the dua...	Two autonomous neural agents communicate iteratively through multi-turn dialectic consensus until 100% agreement verifies the rigorous mathematical proof of truth	true	100	3	7.26	0.001045
sft-cli-run-1788743895488	gpqa-diamond-04	Compton Relativistic Scattering Wavelength Shift at 90-Degree Deflection	GPQA Diamond	science	Quantum Electrodynamics & Compton Scattering	Hard	openrouter/free + qwen/qwen-2.5-coder-32b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A high-energy gamma photon with initial incident wavelength $\lambda = 0.500\text{ pm}$ ($5.00 \times 10^{-13}\text{ m}$) undergoes Compton scattering against an initially stationary free electron ($m_e = 9.109 \times 10^{-31}\text{ kg}$). The scattered photon emerges at a scattering angle of $\theta = 90.0^\circ$ rela...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A high-energy gamma photon wit...	lambda = 2.926 pm, Energy Fraction = 0.829	true	100	2	8.16	0.001236
sft-cli-run-1788743827612	gpqa-diamond-02	Competitive Enzyme Inhibition Apparent Km Ratio	GPQA Diamond	science	Biochemistry & Michaelis-Menten Kinetics	Hard	openrouter/free + qwen/qwen-2.5-72b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	An enzyme follows standard Michaelis-Menten kinetics with substrate $S$, having an uninhibited Michaelis constant $K_m = 4.0\text{ \mu M}$ and $V_{\text{max}} = 100\text{ \mu M/s}$. A competitive inhibitor $I$ with dissociation constant $K_i = 2.0\text{ \mu M}$ is added at a concentration $[I] = 6.0\text{ \mu M}$. In ...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "An enzyme follows standard Mic...	Km_app = 16 μM, v = 50 μM/s	true	100	3	5.61	0.001305
sft-cli-run-1788743654821	frontiermath-03	Ramanujan Expander Graph Alon-Boppana Bound & Second Eigenvalue	FrontierMath	math	Spectral Graph Theory & Expander Graphs	Hard	deepseek/deepseek-chat:free + qwen/qwen-2.5-coder-32b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Let $G$ be an infinite family of connected $d$-regular graphs with adjacency eigenvalues $d = \lambda_1 > \lambda_2 \ge \dots \ge \lambda_n \ge -d$. 1. According to the Alon-Boppana theorem, for any $\epsilon > 0$, as $n \to \infty$, the second largest eigenvalue satisfies $\lambda_2 \ge 2\sqrt{d - 1} - \epsilon$. A $d...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Let $G$ be an infinite family ...	Bound: 6, Spectral Gap: 4	true	100	3	10.53	0.001035
sft-cli-run-1788743327584	abstract-03	Recursive Difference Pattern Induction	ARC Challenge	abstract	Polynomial Sequence Induction	Medium	deepseek/deepseek-r1-distill-llama-70b:free + deepseek/deepseek-chat:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A mathematical sequence is generated by a cubic polynomial rule: $a_1 = 7, \quad a_2 = 26, \quad a_3 = 63, \quad a_4 = 124, \quad a_5 = 215$ Find the underlying algebraic generating formula $a_n$ and compute the exact value of the 10th term ($a_{10}$). State the exact numerical integer value of $a_{10}$.	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A mathematical sequence is gen...	1330	true	100	3	10.05	0.001216
sft-cli-run-1788743108438	strategy-02	Cournot Duopoly Equilibrium Pricing	Game Theory	strategy	Oligopoly Pricing & Nash Equilibrium	Hard	meta-llama/llama-3.3-70b-instruct:free + qwen/qwen-2.5-72b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Two identical competing software firms, Firm A and Firm B, simultaneously choose their output quantities $q_A$ and $q_B$ of an AI API service. The inverse market demand function is given by: $P = 120 - (q_A + q_B)$ Both firms have an identical constant marginal cost of $MC = 12$ per unit and zero fixed costs. Each fir...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Two identical competing softwa...	qA = 36, qB = 36, Price = $48	true	100	2	5.96	0.001277
sft-cli-run-1788742775453	gpqa-diamond-05	Crick Wobble Hypothesis Inosine Base Pairing & tRNA Synthetase Universality	GPQA Diamond	science	Molecular Genetics & Translation Wobble	Hard	mistralai/mistral-7b-instruct:free + qwen/qwen-2.5-72b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In standard biological translation across living organisms: 1. In Francis Crick's classic Wobble Hypothesis, post-transcriptional deamination of adenosine at the first (5') anticodon position produces the non-canonical modified nucleoside Inosine (I). What are ALL three mRNA 3'-codon bases that Inosine in the wobble po...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In standard biological transla...	Wobble Bases: A, U, C, Synthetases: 20	true	100	3	15.33	0.001027
sft-cli-run-1788742648439	gpqa-diamond-01	3D Isotropic Quantum Harmonic Oscillator Degeneracy	GPQA Diamond	science	Quantum Mechanics & Degeneracy	Extreme	mistralai/mistral-7b-instruct:free + qwen/qwen-2.5-72b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	Consider a particle of mass $m$ trapped in a three-dimensional isotropic quantum harmonic oscillator potential $V(r) = \frac{1}{2} m \omega^2 (x^2 + y^2 + z^2)$. The total energy eigenvalues are quantized according to: $$E_n = \hbar \omega \left(n + \frac{3}{2}\right), \quad n = n_x + n_y + n_z \in \{0, 1, 2, 3, \dots\...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "Consider a particle of mass $m...	21 states	true	100	2	10.85	0.001071
sft-cli-run-1788742608528	mmlu-pro-01	Relativistic Carnot Engine Efficiency with Heat Exchange	MMLU-Pro	science	Physics & Thermodynamics	Hard	deepseek/deepseek-r1:free + meta-llama/llama-3.1-8b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A hypothetical relativistic heat engine operates between two thermal reservoirs with temperatures $T_H = 1200\text{ K}$ and $T_C = 300\text{ K}$. During each cycle, the working fluid absorbs $Q_H = 4800\text{ J}$ of heat from the hot reservoir. Due to finite-time heat transfer kinetics across the boundary thermal condu...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A hypothetical relativistic he...	Efficiency = 50%, Work = 2400 J	true	100	2	51.62	0.000618
sft-cli-run-1788742497323	hle-02	TKNN Invariant First Chern Number and Quantized Hall Conductance	Humanity's Last Exam	science	Topological Quantum Matter & Chern Numbers	Extreme	deepseek/deepseek-r1-distill-llama-70b:free + deepseek/deepseek-r1:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	In the integer quantum Hall effect (Thouless-Kohmoto-Nightingale-den Nijs / TKNN formulation), the quantized Hall conductance $\sigma_{xy}$ of an isolated 2D electron gas on a periodic lattice torus is governed by the topological first Chern number $C_1 \in \mathbb{Z}$: $$\sigma_{xy} = C_1 \frac{e^2}{h}, \quad C_1 = \f...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "In the integer quantum Hall ef...	Chern Number: 1, Hall Conductance: 1 e^2/h	true	100	3	5.2	0.001306
sft-cli-run-1788742412819	abstract-03	Recursive Difference Pattern Induction	ARC Challenge	abstract	Polynomial Sequence Induction	Medium	qwen/qwen-2.5-72b-instruct:free + deepseek/deepseek-r1:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	A mathematical sequence is generated by a cubic polynomial rule: $a_1 = 7, \quad a_2 = 26, \quad a_3 = 63, \quad a_4 = 124, \quad a_5 = 215$ Find the underlying algebraic generating formula $a_n$ and compute the exact value of the 10th term ($a_{10}$). State the exact numerical integer value of $a_{10}$.	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "A mathematical sequence is gen...	1330	true	100	2	5.19	0.001456
sft-cli-run-1788742329258	logic-03	The Three Mislabelled Fruit Crates	Formal Logic	logic	Deductive Elimination	Easy	microsoft/phi-3-mini-128k-instruct:free + qwen/qwen-2.5-coder-32b-instruct:free	You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement.	You are in a warehouse with 3 sealed boxes. One contains only Apples, one contains only Oranges, and one contains a mixture of Apples and Oranges. All three boxes are incorrectly labeled: Box 1 says "Apples", Box 2 says "Oranges", Box 3 says "Apples & Oranges". You are allowed to pick ONE box, reach inside blindly, an...	[ { "role": "system", "content": "You are a frontier reasoning model trained via multi-agent double-blind consensus. Solve with strict mathematical, scientific, and algorithmic rigor. Conclude with a bold FINAL ANSWER: statement." }, { "role": "user", "content": "You are in a warehouse with 3 ...	The box labeled "Apples & Oranges" (Box 3)	true	100	3	7.6	0.001073

End of preview. Expand in Data Studio

GlimmaryKarl/DualBlind

Curated Frontier Reasoning and Direct Preference Optimization (DPO) Dataset Generated from Double-Blind Multi-Agent Arena Evaluations.

This dataset was generated using the DualBlind AI Benchmark Arena. In this setup, two independent frontier AI models engage in multi-turn double-blind dialogue to solve extreme-difficulty benchmark problems, verifying their peer's proofs, raising counter-examples, and reaching mathematical consensus.

Dataset Structure & Strict Quality Gate

Strict 100% Accuracy Quality Gate (Active): Only trials that achieved 100% automated verification against canonical ground-truth solutions are admitted into GlimmaryKarl/DualBlind. Incomplete trials, partial solutions (<100%), and unverified outputs are strictly blocked from entry.

Total Arena Trials Evaluated: 3363
Verified 100% Accurate Trials Admitted: 2992 (89% pass rate)
Ineligible Trials Filtered Out (<100% or Refuted): 371
Quality Gate Policy: Zero-tolerance for unverified or imperfect conjectures. Every single sample provides a mathematically and algorithmically verified canonical solution.

Subsets:

sft_reasoning (2962 samples): Supervised Fine-Tuning records containing high-density step-by-step chain-of-thought reasoning traces culminating in 100% verified ground-truth solutions.
dpo_preferences (2961 pairs): Direct Preference Optimization pairs (prompt, chosen, rejected) pairing the 100% verified consensus proof as chosen against flawed peer conjectures or refuted hypotheses as rejected.

Benchmark Suites Covered

GPQA Diamond: 411 runs
MMLU-Pro: 270 runs
FrontierMath: 190 runs
Game Theory: 426 runs
ARC Challenge: 348 runs
SWE-bench: 393 runs
IFEval: 236 runs
Humanity's Last Exam: 242 runs
Formal Logic: 310 runs
MATH / AIME: 143 runs
General Benchmark: 23 runs

Models Represented

meta-llama/llama-3.3-70b-instruct:free
qwen/qwq-32b:free
openrouter/free
gemini-1.5-flash
qwen/qwen-2.5-coder-32b-instruct:free
gemini-2.5-flash
qwen/qwen-2.5-72b-instruct:free
deepseek/deepseek-r1:free
microsoft/phi-3-mini-128k-instruct:free
deepseek/deepseek-chat:free
google/gemini-2.0-flash-exp:free
deepseek/deepseek-r1-distill-llama-70b:free
meta-llama/llama-3.1-8b-instruct:free
mistralai/mistral-7b-instruct:free
gemini-2.0-flash
poolside/laguna-s-2.1:free
poolside/laguna-xs-2.1:free
nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free
z-ai/glm-5.2:free
nvidia/nemotron-3.5-lightning:free
nvidia/nemotron-3-ultra-550b-a55b:free
meta-llama/llama-3.2-3b-instruct:free
mistralai/mistral-small-24b-instruct-2501:free
cognitivecomputations/dolphin3.0-r1-mistral-24b:free
gemini-3.7-flash
google/gemma-2-9b-it:free
nvidia/llama-3.1-nemotron-70b-instruct:free
openai/gpt-3.5-turbo
thedrummer/skyfall-36b-v2
liquid/lfm-2.5-2.6b:free
google/gemma-4-26b-a4b-it:free
minimax/minimax-m3:free
z-ai/glm-4.7-flash
tencent/hy-mt2-30b-a3b
openai/gpt-4o-mini
meta-llama/llama-3.3-70b-instruct
qwen/qwen-2.5-72b-instruct
meta/muse-glimmer-30b
google-gemini-3.7-flash
claude-3-7-sonnet-20250219
deepseek-r1
o3-mini
gpt-4.5-preview
ox-alpha
nvidia/nemotron-3-nano
amazon/nova-lite-v1
meta-llama/llama-3.1-70b-instruct
deepseek-deepseek-v3
anthropic-claude-3-haiku
gpt-4o

Usage with Hugging Face Datasets

from datasets import load_dataset

# Load SFT Reasoning split
sft_ds = load_dataset("GlimmaryKarl/DualBlind", "sft_reasoning", split="train")
print(sft_ds[0]["messages"])

# Load DPO Preference split
dpo_ds = load_dataset("GlimmaryKarl/DualBlind", "dpo_preferences", split="train")
print(dpo_ds[0]["chosen"])
print(dpo_ds[0]["rejected"])

100% Free Fine-Tuning Guide (Unsloth on Google Colab T4)

You can fine-tune Llama-3.1-8B-Instruct or Qwen-2.5-7B in ~25 minutes on a free Google Colab T4 GPU:

pip install "unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git"
pip install --no-deps trl peft accelerate bitsandbytes

from unsloth import FastLanguageModel
from trl import SFTTrainer
from transformers import TrainingArguments
from datasets import load_dataset

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="unsloth/Llama-3.1-8B-Instruct",
    max_seq_length=2048,
    load_in_4bit=True,
)

model = FastLanguageModel.get_peft_model(
    model,
    r=16,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
    lora_alpha=16,
    lora_dropout=0,
    bias="none",
)

dataset = load_dataset("GlimmaryKarl/DualBlind", "sft_reasoning", split="train")

trainer = SFTTrainer(
    model=model,
    train_dataset=dataset,
    dataset_text_field="text",
    max_seq_length=2048,
    tokenizer=tokenizer,
    args=TrainingArguments(
        per_device_train_batch_size=2,
        gradient_accumulation_steps=4,
        warmup_steps=5,
        max_steps=60,
        learning_rate=2e-4,
        fp16=True,
        logging_steps=1,
        output_dir="outputs",
    ),
)
trainer.train()

License

Apache 2.0. Free for commercial, research, and educational use.

Downloads last month: 60