API reference & guides

Classical physics toolkit: mechanics, kinematics, thermo, EM, optics, relativity, integrators, NL query.

Astronomy toolkit: coordinates, observing, orbits, cosmology, catalogs (optional C++ path).

Classic computer vision on NumPy arrays: I/O, transforms, color, filters, draw, metrics, OpenCV ops.

Caching with TTL: MemoryCache, DiskCache, LLMCache, and @cached decorator.

CSV/JSON/JSONL loaders, splits, text augmentation, and schema validation.

Built-in benchmarks, generators, and dataset file I/O helpers.

BPETokenizer, WordPieceTokenizer, and Vocabulary.

Step-based pipelines with retry, fallback, and timing.

SQLite key-value store, persistent vector store, and chat history.

Experiment tracking with Tracker/Run and compare helpers.

Similarity, faithfulness, toxicity/PII checks, and cost tracking.

Trie, Bloom filter, LRU, heaps, priority queue, Union-Find.

FastAPI chat/RAG/health server helpers (pip install aqwel-aion[serve]).

NumPy-first scalers, encoders, imputers, and feature transforms.

Classical ML estimators: regression, classification, clustering, PCA, pipelines.

Classification, regression, clustering, ranking, and NLP metrics.

Grid/random/Bayesian search, cross-validation, early stopping.

Lightweight HTML/VDOM components and optional Gradio/Streamlit launchers.

Unified SQLite plus optional MySQL/Postgres/Mongo/Redis backends.

Experiment runner, BenchmarkSuite, and manifest helpers.

Token and cost dashboard helpers (aion usage).

CPU/RAM/disk/GPU snapshots and runtime tracking.

Chunk statistics, rolling ops, histogram, prefix sum (optional native).

Environment and optional-dependency health checks (aion doctor).

Local Hub dashboard for library info, modules, and snippet runs.

Timing and reproducibility helpers: seeds, dataset snapshots, baseline suite.

Mathematical primitives for neural simulation, linear algebra, and statistical analysis.

Binary, jump, exponential, linear search; bounds; first/last occurrence; rotated and interpolation search.

Flatten, chunk, sliding window, rolling sum, dedupe, pad, pairwise.

BFS, DFS, topological sort, Dijkstra, connected components, unweighted shortest path.

Array, histogram, scatter, running mean, boxplot, density, CDF, error bars, dual axis.

Heatmap, confusion matrix, surface, contour, correlation, attention, sparsity.

Training history, metrics, train vs val, learning rate, early stopping.

finalize_plot, save_plot, close_figure.

3D plots and figure export helpers (v0.1.9).

Text embeddings and vector similarity (optional: sentence-transformers).

Classification and regression metrics; file-based evaluation.

Prompt templates and utilities for research workflows.

General utilities: format_bytes, slugify, UUID, validation.

Real-time file and directory change monitoring.

Git repository operations (optional: GitPython).

Language detection and code parsing for 30+ languages.

Extract comments, function names, and class names from code.

NLP utilities for cleaning, extraction, and analysis.

File and directory operations, hashing, search, organization.

Code complexity, explanation, extraction, and quality analysis.

API docs, user guide, changelog, and report generation (optional: ReportLab).

Optional C++ extension: fast reductions, activations, bounds on 1D arrays (NumPy fallback).

Streaming reads, atomic writes, checksums for safe file handling.

Remote LLM clients: OpenAI, Gemini, Anthropic, OpenAI-compatible servers.

Tool schemas, registry, multi-turn tool loop, retry, token estimates.

Chunking, in-memory/FAISS stores, simple RAG index pipeline.

TOML/YAML loading, merging, nested keys, env overrides.

Package root (`aion.former`): re-exports models, training, datasets, and visualization for a NumPy autograd decoder-only transformer.

Tensor and differentiable ops (`aion.former.core`): autograd, matmul, softmax, layer_norm, relu.

Transformer stack and layers (`aion.former.models`): blocks, attention, FFN, embeddings.

Trainer, Adam, loss, and .npz checkpoint helpers (`aion.former.training`).

TextDataset, tokenizer, and create_dataloader (`aion.former.datasets`).

Attention and training plots (`aion.former.visualization`; requires matplotlib).

Command-line interface: chat, embed, eval, prompts, watch, git.