Expand description
Machine Learning Library - Edge AI and Neural Network Computing
This module provides high-performance machine learning operations leveraging Phase 2 enhancements:
- NVMe Computational Storage (CSD) for hardware-accelerated neural computations
- Ambient Sub-Threshold Orchestration for mobile edge AI optimization
- Hardware-Sympathetic Storage (ZNS) for zero-copy model storage
- Zero-Copy LoRA Multiplexing for efficient model serving
This module was split out of the former monolithic machine_learning.rs into a
subdirectory library (pure code motion; no behaviour, logic, or signature changes).
The public surface is re-exported unchanged so every external path
(crate::specialized_libs::machine_learning::<Item>) resolves exactly as before.
Structs§
- Augmentation
Step - Augmentation step
- Auto
Tuner - Auto tuner
- Backend
Capabilities - Backend capabilities
- Backend
Metrics - Backend metrics
- Batch
Optimization Metrics - Batch optimization metrics
- Batch
Optimizer - Batch optimizer
- Batch
Processor - Batch processor
- Bottleneck
Detector - Bottleneck detector
- Cache
Entry - Cache entry
- Cache
Policy - Cache policy
- Cache
Stats - Cache statistics
- Completed
Request - Completed request
- Compression
Quality Metrics - Compression quality metrics
- Compression
Statistics - Compression statistics
- Conversion
Pipeline - Conversion pipeline
- Conversion
Step - Conversion step
- Data
Augmenter - Data augmenters
- Data
Loader - Data loaders
- Data
Pipeline - Data pipeline
- Data
Source - Data sources
- Data
Transformer - Data transformers
- Detection
Thresholds - Detection thresholds
- Distillation
Config - Configuration for the teacher-to-student loop supported by this module.
- Distillation
Report - Evidence emitted by a completed teacher-to-student training run.
- Early
Stopping - Early stopping
- Format
Converter - Format converter
- Hardware
Spec - Hardware specifications
- Health
Check - Health check
- Health
Checker - Health checker
- Hyperparameter
- Hyperparameter
- Hyperparameter
Constraint - Hyperparameter constraints
- Hyperparameter
Range - Hyperparameter range
- Hyperparameter
Tuner - Hyperparameter tuner
- Inference
Backend - Inference backends
- Inference
Engine - Inference engine
- Inference
Metrics - Inference metrics
- Inference
Optimizer - Inference optimizer
- Inference
Parameters - Inference parameters
- Inference
Request - Inference request
- Inference
Result - Inference result
- Layer
Connection - Layer connection
- Layer
Info - Layer information
- Load
Balancer - Load balancer
- Loading
Cache - Loading cache
- Loading
Parameters - Loading parameters
- Loading
Strategy - Loading strategies
- MLOperation
Result - ML operation result
- MLOptimization
Engine - ML optimization engine
- MLPerformance
Metrics - ML library performance summary metrics
- MLPerformance
Monitor - ML performance monitor
- Machine
Learning Library - Machine Learning Library Manager
- Model
- Model representation
- Model
Architecture - Model architecture
- Model
Branch - Model branch
- Model
Cache - Model cache
- Model
Cache Entry - Model cache entry
- Model
Cache Policy - Model cache policy
- Model
Cache Stats - Model cache statistics
- Model
Catalog - Model catalog for model management
- Model
Change - Model change
- Model
Compression - Model compression
- Model
Converter - Model converter
- Model
Index Entry - Model index entry
- Model
Loader - Model loader
- Model
Manager - Model manager for neural network models
- Model
Metadata - Model metadata
- Model
Metrics - Model metrics
- Model
Parameters - Model parameters
- Model
Performance - Model performance metrics
- Model
Relationship - Model relationships
- Model
Search Engine - Model search engine
- Model
Search Index - Model search index
- Model
Storage - Model storage using ZNS for efficient model storage
- Model
Tag - Model tag
- Model
Version - Model version
- Model
Version Control - Model version control
- Model
Zone - Model zone for different model types
- Optimization
Constraint - Optimization constraints
- Optimization
Objective - Optimization objectives
- Optimization
Parameters - Optimization parameters
- Optimization
Strategy - Optimization strategies
- Performance
Analyzer - Performance analyzer
- Performance
Characteristics - Performance characteristics
- Performance
Metrics - Performance metrics for inference and optimization results
- Performance
Profile - Performance profile
- Progress
Metrics - Progress metrics
- Progress
Tracker - Progress tracker
- Pruning
Report - Measured result of an exact magnitude-pruning pass.
- Quality
Assurance - Quality assurance
- Quantization
Parameters - Calibration parameters required to reconstruct a PTQ tensor.
- Quantization
Report - Measured result of a real post-training quantization pass.
- Queue
Manager - Queue manager
- Request
Scheduler - Request scheduler
- Resource
- Resource
- Resource
Manager - Resource manager
- Resource
Utilization - Resource utilization
- Result
Metadata - Result metadata
- Running
Request - Running request
- Software
Spec - Software specifications
- Stopping
Criteria - Stopping criteria
- System
Metrics - System metrics
- System
Training Metrics - System-wide training metrics
- Test
Case - Test cases
- Test
Configuration - Test configuration
- Test
Environment - Test environment
- Test
Metrics - Test metrics
- Test
Result - Test result
- Test
Results - Test results
- Test
Suite - Test suite
- Test
Summary - Test summary
- Training
Backend - Training backends
- Training
Capabilities - Training capabilities
- Training
Config - Training configuration
- Training
Engine - Training engine
- Training
Job - Training job representation
- Training
Metrics - Training metrics
- Training
Optimizer - Training optimizer
- Training
Result - Result of a completed training run.
- Training
Scheduler - Training scheduler
- Transformation
Step - A single data transformation step in a pipeline
- Tuning
Configuration - Tuning configuration
- Tuning
History - Tuning history
- Tuning
Objective - Tuning objectives
- Tuning
Record - Tuning record
- Tuning
Space - Tuning space
- Utilization
Record - Utilization record
- Utilization
Tracker - Utilization tracker
- Validation
Condition - Validation conditions
- Validation
Engine - Validation engine
- Validation
Logic - Validation logic
- Validation
Rule - Validation rules
- Validator
- Validator
Enums§
- Access
Pattern - Access patterns for optimization
- Activation
Function - Activation functions
- Allocation
Strategy - Allocation strategies
- Analysis
Method - Analysis methods
- Augmentation
Step Type - Augmentation step types
- Availability
- Availability
- Batch
Optimization Algorithm - Batch optimization algorithms
- Batching
Strategy - Batching strategies
- Bottleneck
Detection Algorithm - Bottleneck detection algorithms
- Change
Type - Change types
- Comparison
Operator - Comparison operators
- Compression
Algorithm - Compression algorithms
- Connection
Type - Connection types
- Constraint
Type - Constraint types
- Conversion
Step Type - Conversion step types
- Data
Augmenter Type - Data augmenter types
- Data
Format - Data formats
- Data
Loader Type - Data loader types
- Data
Source Type - Data source types
- Data
Transformer Type - Data transformer types
- Eviction
Policy - Cache eviction policy
- Health
Check Type - Health check types
- Hyperparameter
Type - Hyperparameter types
- Indexing
Strategy - Indexing strategies
- Inference
Backend Type - Inference backend types
- Inference
Optimization Strategy - Inference optimization strategies
- Layer
Type - Layer types
- Load
Balancing Strategy - Load balancing strategies
- Loading
Strategy Type - Loading strategy types
- MLError
- ML error types
- MLFramework
- ML frameworks
- MLOptimization
Algorithm - ML optimization algorithms
- Model
Eviction Policy - Model eviction policies
- Model
Relationship Type - Model relationship types
- Model
Type - Model types
- Model
Zone Type - Model zone types
- Objective
Type - Objective types
- Optimization
Level - Optimization levels
- Optimization
Strategy Type - Optimization strategy types
- Precision
- Precision types
- Priority
Level - Priority levels
- Quantization
Scheme - Model-agnostic post-training quantization schemes.
- Request
Priority - Request priorities
- Resource
Type - Resource types
- Scheduling
Policy - Scheduling policies
- Search
Engine Type - Search engine types
- Stopping
Mode - Stopping modes
- Test
Type - Test types
- Training
Algorithm - Training algorithms
- Training
Backend Type - Training backend types
- Training
Optimization Algorithm - Training optimization algorithms
- Training
Scheduling Policy - Training scheduling policies
- Training
Status - Training status
- Tuning
Algorithm - Tuning algorithms
- Validation
Action - Validation actions
- Validation
Rule Type - Validation rule types
- Validation
Value - Validation values
- Validator
Type - Validator types
Constants§
- GGUF_
EMBEDDING_ PREVIEW_ TOKENS - Maximum number of token embeddings materialised into
Model.weightswhen loading a real GGUF file. The full vocabulary embedding table can be multiple gigabytes, so only a bounded preview is kept in the in-memoryVec<f64>(this is not a hot-path module). - PRUNING_
MASK_ BITS_ PER_ BYTE - Number of logical pruning decisions encoded in one mask byte.