ruvllm 2.2.0

LLM serving runtime with Ruvector integration - Paged attention, KV cache, and SONA learning
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48

//! RuvLLM Evaluation Harness
//!
//! Three-layer evaluation framework:
//! 1. **Correctness**: Does the patch actually work?
//! 2. **Diff Quality**: Does it behave like a senior engineer?
//! 3. **Systems Economics**: Is it worth running at scale?
//!
//! ## Usage
//!
//! ```rust,ignore
//! use ruvllm::evaluation::{EvaluationHarness, EvalConfig, AblationMode};
//!
//! let config = EvalConfig {
//!     task_count: 100,
//!     seeds: vec![42, 123, 456],
//!     ablation_modes: vec![
//!         AblationMode::Baseline,
//!         AblationMode::RetrievalOnly,
//!         AblationMode::AdaptersOnly,
//!         AblationMode::RetrievalPlusAdapters,
//!         AblationMode::Full, // retrieval + adapters + SONA
//!     ],
//!     ..Default::default()
//! };
//!
//! let harness = EvaluationHarness::new(config);
//! let report = harness.run_evaluation(&tasks).await?;
//! println!("{}", report.to_leaderboard());
//! ```

mod correctness;
mod diff_quality;
mod economics;
mod harness;
mod metrics;
mod real_harness;
mod report;
pub mod swe_bench;

pub use correctness::{CorrectnessMetrics, TaskResult, TestSuiteResult, VerificationLevel};
pub use diff_quality::{DiffAnalyzer, DiffQualityMetrics, EditLocality, Minimality, ReviewBurden};
pub use economics::{CostTracker, EconomicsMetrics, LatencyDistribution, StabilityMetrics};
pub use harness::{
    AblationMode, EvalConfig, EvalReport, EvalRun, EvalTask, EvaluationHarness, ModeMetrics,
};
pub use metrics::{AggregatedMetrics, MetricCollector, MetricSnapshot};
pub use real_harness::{RealEvaluationHarness, RealInferenceConfig};
pub use report::{AblationComparison, LeaderboardEntry};