Skip to main content

Module experiment

batuta

Module experiment 

Source
Expand description

Experiment Tracking Integration Module

Integrates with Entrenar Experiment Tracking Spec v1.8.0 for orchestrating ML experiment workflows with full traceability, cost optimization, and academic research support.

§Entrenar CLI (v0.2.4)

The entrenar crate provides a comprehensive CLI:

# Training (YAML Mode v1.0)
entrenar train config.yaml           # Train from declarative YAML
entrenar validate config.yaml        # Validate configuration
entrenar init --template lora        # Generate config template

# Model Operations
entrenar quantize model.safetensors --bits 4
entrenar merge model1.st model2.st --method ties

# Research Workflows
entrenar research init --id my-dataset
entrenar research cite artifact.yaml --format bibtex

# Inspection & Auditing
entrenar inspect model.safetensors   # Model/data inspection
entrenar audit data.parquet --type bias
entrenar monitor data.parquet        # Drift detection

# Benchmarking (entrenar-bench)
entrenar-bench temperature --start 1.0 --end 8.0
entrenar-bench cost-performance --gpu a100-80gb

§MCP Tooling (pmcp v1.8.6 + pforge v0.1.4)

The stack includes Model Context Protocol (MCP) infrastructure:

# pmcp - Rust SDK for MCP servers/clients
# Build MCP servers with full TypeScript SDK compatibility

# pforge - Declarative MCP framework
pforge new my-server              # Create new MCP server project
pforge serve                       # Run MCP server

# Define tools in YAML (pforge.yaml):
# tools:
#   - type: native
#     name: train_model
#     handler: { path: handlers::train }
#     params:
#       config: { type: string, required: true }

Handler Types:

  • native - Rust functions with full type safety
  • cli - Execute shell commands
  • http - Proxy HTTP endpoints
  • pipeline - Chain multiple tools together

§Features

  • ComputeDevice abstraction (CPU/GPU/TPU/AppleSilicon)
  • EnergyMetrics and CostMetrics for efficiency tracking
  • ModelParadigm classification
  • CostPerformanceBenchmark with Pareto frontier analysis
  • SovereignDistribution for air-gapped deployments
  • ResearchArtifact with ORCID/CRediT academic support
  • CitationMetadata for BibTeX/CFF generation
  • Experiment tree visualization for run comparison (MLflow replacement)
  • YAML Mode Training v1.0 declarative configuration

Re-exports§

pub use types::AppleChip;
pub use types::ComputeDevice;
pub use types::ComputeIntensity;
pub use types::CpuArchitecture;
pub use types::ExperimentError;
pub use types::GpuVendor;
pub use types::ModelParadigm;
pub use types::PlatformEfficiency;
pub use types::TpuVersion;
pub use metrics::CostMetrics;
pub use metrics::EnergyMetrics;
pub use benchmark::CostPerformanceBenchmark;
pub use benchmark::CostPerformancePoint;
pub use research::CitationMetadata;
pub use research::CitationType;
pub use research::CreditRole;
pub use research::Orcid;
pub use research::PreRegistration;
pub use research::ResearchArtifact;
pub use research::ResearchContributor;
pub use run::ExperimentRun;
pub use run::ExperimentStorage;
pub use run::InMemoryExperimentStorage;
pub use run::RunStatus;
pub use sovereign::ArtifactSignature;
pub use sovereign::ArtifactType;
pub use sovereign::OfflineRegistryConfig;
pub use sovereign::SignatureAlgorithm;
pub use sovereign::SovereignArtifact;
pub use sovereign::SovereignDistribution;

Modules§

benchmark
Cost-performance benchmarking with Pareto frontier analysis.
metrics
Energy and cost metrics for experiment tracking.
research
Academic research support with ORCID, CRediT, and citation generation.
run
Experiment run tracking and storage.
sovereign
Sovereign distribution manifest for air-gapped deployments
tree
Experiment Tracking Frameworks Tree Visualization
types
Core type definitions for experiment tracking.