Expand description
Training utilities and checkpoint management
This module provides:
ttt: Test-Time Training for research-valid model adaptationtenant_delta: Per-tenant LoRA delta for isolated TTTdelta_pool: Per-tenant delta registry with LRU evictionmerge: LoRA adapter merging strategies (DO-Merge, additive, replace) — 1-vs-1 save-time pathdelta_reducer: order-independent N-way param-server reducer (stack + truncated SVD)quality_filter: Heuristic-based output filtering (NOT for training)checkpoint: Checkpoint management with git integrationdata_loader: Training data loading utilities
Re-exports§
pub use adaptation_state::AdaptationStrategy;pub use adaptation_state::DeltaAdaptationState;pub use adaptation_state::EvictedSnapshot;pub use adaptation_state::GuardStatus;pub use adaptation_state::ResolveOutcome;pub use adaptation_state::WritebackInfo;pub use checkpoint::CheckpointConfig;pub use checkpoint::CheckpointInfo;pub use checkpoint::CheckpointManager;pub use checkpoint::CheckpointRequest;pub use checkpoint::TrainingMetrics;pub use checkpoint::WeightFormat;pub use checkpoint::WeightSnapshot;pub use data_loader::ChatTemplateDataLoader;pub use data_loader::TrainingDataset;pub use data_loader::TrainingSample;pub use delta_pool::DeltaPool;pub use quality_filter::FilterReason;pub use quality_filter::FilterResult;pub use quality_filter::QualityFilter;pub use quality_filter::QualityFilterConfig;pub use tenant_delta::load_state_dict_from_bytes;pub use tenant_delta::serialize_state_dict_to_bytes;pub use tenant_delta::TenantDelta;pub use tenant_delta::TenantDeltaConfig;pub use merge::merge_state_dicts;pub use merge::MergeStrategy;pub use delta_reducer::rank_cap_for_dtype_bytes;pub use delta_reducer::reduce_state_dicts;pub use delta_reducer::Contribution;pub use delta_reducer::ContributionGate;pub use delta_reducer::ContributorState;pub use delta_reducer::DeltaReducer;pub use delta_reducer::NoopValidation;pub use delta_reducer::Provenance;pub use delta_reducer::StackSvdReducer;pub use delta_reducer::ValidationHook;pub use ttt::GradientGatingConfig;pub use ttt::RankOracle;pub use ttt::RankOracleConfig;pub use ttt::TTTConfig;pub use ttt::TTTContext;pub use ttt::TTTOverrides;pub use ttt::TTTResult;pub use ttt::TTTVerifier;pub use ttt::TestTimeTrainer;pub use pipeline::compute_ntp_loss_split;pub use pipeline::forward_train_pipeline;pub use pipeline::stage_ranges;pub use pipeline::TrainStage;
Modules§
- adaptation_
state - checkpoint
- Async checkpoint management for training
- data_
loader - Training data loading and processing for LoRA fine-tuning
- delta_
pool - Per-tenant delta registry with LRU eviction
- delta_
reducer - N-way delta aggregation (param-server reducer) — replaces DO-Merge for multi-contributor merges.
- merge
- LoRA adapter merging strategies
- muon
- Muon optimizer: MomentUm Orthogonalized by Newton-Schulz.
- pipeline
- Pipeline-split (2b) training orchestration for TTT-on-split (#316).
- quality_
filter - Quality Filter Module
- tenant_
delta - Per-tenant LoRA delta for isolated Test-Time Training
- ttt
- Test-Time Training (TTT) Module