Expand description
Adaptive ML inference server.
This crate provides the core functionality for:
- Real-time weight updates for neural networks
- Dynamic adaptive ML inference
- Hardware-accelerated storage
- Memory-mapped disk persistence
- FlightSQL interface for embeddings and similarity search
Re-exports§
pub use events::EventEnvelope;pub use events::EventPayload;pub use events::EventSource;
Modules§
- account
- Account identity — host-form
did:webaccount zone, DNS interface, and DNS-01 wildcard TLS (epic #1158, Phase A3 / ticket #1162). - annotations_
capnp - api
- REST API for creating and managing sparse auto-regressive LoRA training layers
- archetypes
- Repository archetype detection and capability system
- auth
- Authorization module for RBAC/ABAC access control
- cli
- Command-line interface module.
- common_
capnp - config
- Unified configuration system for Hyprstream
- constants
- Shared constants for the hyprstream application
- error
- events
- Event types for hyprstream
- git
- Git operations for model version control
- inference
- Inference service abstraction for in-process and future IPC access.
- mac
- Mandatory Access Control — the Policy Decision Point (PDP) for the 9P/VFS surface.
- nine_
capnp - optional_
capnp - runtime
- Runtime abstraction layer for inference engines
- schema
- Cap’n Proto generated schemas
- server
- Hyprstream server implementation with OpenAI-compatible API
- services
- Service layer for hyprstream
- storage
- Storage for model weights and adapters
- streaming_
capnp - systemd
- Systemd integration module
- training
- Training utilities and checkpoint management
- tui
- TUI display server — terminal multiplexer over RPC.
Structs§
- Generation
Result - Generation result with metadata
- Runtime
Config - Runtime execution configuration
- Torch
Engine - PyTorch inference engine using tch-rs
Enums§
- Finish
Reason - Why generation stopped
Traits§
- Runtime
Engine - Core runtime engine trait - all engines implement this
Functions§
- init
- Create the default runtime engine (now uses PyTorch)