Reasoning

41 public types in this namespace, organized by kind.

Models & Types (33)

Type Summary
ARCAGIBenchmark<T> ARC-AGI (Abstract Reasoning Corpus - Artificial General Intelligence) benchmark.
BenchmarkProblem Represents a single problem in a benchmark dataset.
BenchmarkResult<T> Results from evaluating a reasoning system on a benchmark.
BoolQBenchmark<T> BoolQ benchmark for evaluating yes/no question answering.
ChainOfThoughtStrategy<T> Implements Chain-of-Thought (CoT) reasoning that solves problems through explicit step-by-step thinking.
CodeReasoner<T> Specialized reasoner for code-related problems including generation, debugging, and explanation.
CodeXGlueBenchmark<T> CodeXGLUE benchmark harness (dataset-loader + metric computation).
CodeXGlueProblem A single CodeXGLUE problem record (task-agnostic).
CommonsenseQABenchmark<T> CommonsenseQA benchmark for evaluating commonsense knowledge and reasoning.
DROPBenchmark<T> DROP (Discrete Reasoning Over Paragraphs) benchmark for numerical and discrete reasoning.
GSM8KBenchmark<T> Grade School Math 8K (GSM8K) benchmark for evaluating mathematical reasoning.
GSM8KDataLoader Loader for GSM8K benchmark dataset.
GSM8KProblem
HellaSwagBenchmark<T> HellaSwag benchmark for evaluating commonsense natural language inference.
HumanEvalBenchmark<T> HumanEval benchmark for evaluating Python code generation capabilities.
HumanEvalDataLoader Loader for HumanEval benchmark dataset.
HumanEvalProblem
LogiQABenchmark<T> LogiQA benchmark for evaluating logical reasoning abilities.
LogicalReasoner<T> Specialized reasoner for formal logic and logical reasoning problems.
MATHBenchmark<T> MATH benchmark for evaluating advanced mathematical reasoning.
MBPPBenchmark<T> MBPP (Mostly Basic Python Problems) benchmark for evaluating Python code generation.
MMLUBenchmark<T> MMLU (Massive Multitask Language Understanding) benchmark for evaluating world knowledge.
MathematicalReasoner<T> Specialized reasoner for mathematical problems using verified reasoning and external verification.
PIQABenchmark<T> PIQA (Physical Interaction Question Answering) benchmark for physical commonsense reasoning.
ProblemEvaluation<T> Result for a single problem evaluation.
ReasoningChain<T> Represents a complete chain of reasoning steps from problem to solution.
ReasoningResult<T> Represents the complete result of a reasoning process, including the answer, reasoning chain, and performance metrics.
ReasoningStep<T> Represents a single step in a reasoning chain, capturing the thought process and evaluation.
ScientificReasoner<T> Specialized reasoner for scientific problems and hypotheses.
SelfConsistencyStrategy<T> Implements Self-Consistency reasoning by sampling multiple reasoning paths and using majority voting.
TreeOfThoughtsStrategy<T> Implements Tree-of-Thoughts (ToT) reasoning that explores multiple reasoning paths in a tree structure.
TruthfulQABenchmark<T> TruthfulQA benchmark for evaluating truthfulness and resistance to falsehoods.
WinoGrandeBenchmark<T> WinoGrande benchmark for evaluating commonsense reasoning through pronoun resolution.

Base Classes (1)

Type Summary
ReasoningStrategyBase<T> Abstract base class for reasoning strategies that solve problems through structured thinking.

Enums (2)

Type Summary
ReasoningMode Available reasoning modes that determine how problems are solved.
SearchAlgorithmType Types of search algorithms available for Tree-of-Thoughts.

Options & Configuration (3)

Type Summary
CodeXGlueBenchmarkOptions Options for configuring CodeXGLUE dataset loading.
HumanEvalBenchmarkOptions
ReasoningConfig Configuration options for reasoning strategies that control how problems are solved.

Helpers & Utilities (2)

Type Summary
CodeXGlueDataLoader Loads CodeXGLUE-style datasets from JSONL (one JSON object per line).
ThoughtNode<T> Represents a node in a tree of thoughts, used for exploring multiple reasoning paths.