Keyboard shortcuts

Press or to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Appendix B: Stack Version Matrix

Last verified: 2026-03-02

ComponentVersionRole in Albor
aprender (apr)0.4.10 (7c27c2b3)Unified CLI: train, tokenize, eval, distill, merge, export, publish, pipeline
entrenar0.7.5 (with local patches: ALB-038/041/043/044 fixes)Training engine, autograd, CudaTransformerTrainer, optimizers, LoRA
trueno0.16.1SIMD/GPU tensor backend
realizar0.8.0Inference engine (SafeTensors loading, teacher model, eval, serving)
alimentar0.2.6Data pipeline, Parquet I/O, HF Hub import, FIM transforms, mixing
repartir2.0.3Distributed compute (future: gradient sync)
forjar1.0.0Pipeline orchestration (DAG engine, infra + task resources)
presentar0.3.2Training visualization (TUI dashboards, WASM, experiment browser)
bashrs (Rash)6.65.0Makefile lint/purify/classify, shell safety, pipeline command validation
batuta0.7.2Stack orchestration, oracle, falsification (108 checks), playbook DAG engine
provable-contracts (pv)0.1.0Design-by-contract YAML specs, Kani proofs, falsification tests
pmat3.6.1TDG scoring, comply check, fault patterns, coverage gaps
certezalatestThree-tier test effectiveness (unit → property → formal)
renacerlatestTracing infrastructure (BrickTracer, spans, metric events)

Note: apr uses [patch.crates-io] to override entrenar/realizar with local paths. The installed entrenar 0.7.5 includes unpublished fixes for ALB-038, ALB-041, ALB-043, ALB-044 (gradient flow, buffer sizes, activation clipping, optimizer hyperparams).