aphelion-framework 1.2.6

Python bindings for the Aphelion AI Framework - unified frontend for AI model development

aphelion-framework

Python bindings for the Aphelion AI Framework.

Installation

pip install aphelion-framework

Building from source

Requires maturin:

pip install maturin
cd crates/aphelion-python
maturin develop

Feature flags

Build with specific backends:

maturin develop --features burn
maturin develop --features cubecl
maturin develop --features tritter-accel
maturin develop --features full  # all backends

Quick Start

import aphelion
import asyncio

async def main():
    # Create configuration
    config = aphelion.ModelConfig("my-model", "1.0.0")
    config.with_param("hidden_size", 256)
    config.with_param("layers", 4)

    # Build graph
    graph = aphelion.BuildGraph()
    node1 = graph.add_node("input", config)
    node2 = graph.add_node("output", config)
    graph.add_edge(node1, node2)

    # Validate
    assert not graph.has_cycle()
    print(f"Hash: {graph.stable_hash()}")

    # Create pipeline context
    backend = aphelion.NullBackend.cpu()
    trace = aphelion.InMemoryTraceSink()
    ctx = aphelion.BuildContext(backend, trace)

    # Execute pipeline (async)
    pipeline = aphelion.BuildPipeline.standard()
    result = await pipeline.execute_async(ctx, graph)

    # Get trace events
    for event in trace.events():
        print(f"[{event.level}] {event.id}: {event.message}")

asyncio.run(main())

API Overview

Configuration

• ModelConfig - Model configuration with typed parameters
• ModelConfigBuilder - Builder pattern for configs

Graph

• NodeId - Unique node identifier
• GraphNode - Graph node with config and metadata
• BuildGraph - Computation graph with cycle detection and hashing

Backend

• NullBackend - CPU-only reference backend for testing

Note: Additional backends (Burn, CubeCL, Tritter-Accel) will be available in future releases. Check HAS_BURN, HAS_CUBECL, HAS_TRITTER_ACCEL feature flags at runtime.

Pipeline

• BuildPipeline - Multi-stage build pipeline with hooks
• BuildContext - Execution context with backend and tracing
• PipelineStage - Custom stage protocol (subclass in Python)

Diagnostics

• TraceLevel - Log levels (Debug, Info, Warn, Error)
• TraceEvent - Structured trace event
• InMemoryTraceSink - Collects events in memory

Validation

• ValidationError - Validation failure details
• NameValidator - Validates model names
• VersionValidator - Validates semver versions
• CompositeValidator - Combines multiple validators

Feature Flags

Python constants to check available backends at runtime:

• HAS_BURN - Whether Burn backend is available
• HAS_CUBECL - Whether CubeCL backend is available
• HAS_RUST_AI_CORE - Whether rust-ai-core is available
• HAS_TRITTER_ACCEL - Whether Tritter hardware acceleration is available

License

MIT OR Apache-2.0