Crate burn_dragon_language 

Language training and inference adapters over the shared Dragon BDH core.

Paper mapping:

• burn_dragon_core::BDH owns the paper-faithful x_neuron, y_gate, y_neuron, and per-layer recurrent rho contract
• this crate layers tokenization, datasets, generation, and training schedules on top of that core without redefining the recurrent state semantics

Re-exports

pub use bitnet_artifact::BITNET_ARTIFACT_BINARY_MAGIC;

pub use bitnet_artifact::LanguageBitNetArtifactBundle;

pub use bitnet_artifact::deserialize_bitnet_artifact_binary;

pub use bitnet_artifact::serialize_bitnet_artifact_binary;

pub use config::ContextStrategyConfig;

pub use config::GenerationConfig;

pub use config::GenerationOutputFormat;

pub use config::GenerationTokenizerSourceConfig;

pub use config::ModelOverrides;

pub use generation::ContextStrategy;

pub use generation::GenerationProfileSnapshot;

pub use generation::GenerationSettings;

pub use generation::generate_text;

pub use generation::generate_tokens;

pub use generation::generate_tokens_chunked;

pub use generation::generation_profile_reset;

pub use generation::generation_profile_snapshot;

pub use generation::prefill_state;

pub use generation::resolve_context_strategy;

pub use generation::sample_next_token;

pub use inference::WgpuFusedCoreOverride;

pub use inference::apply_wgpu_fused_core_override;

pub use inference::build_model_config;

pub use inference::build_model_config_with_tokenizer;

pub use inference::is_wgpu_backend_name;

pub use loss::language_model_loss;

pub use summary_events::resolve_summary_memory_write_triggers;

pub use summary_events::summary_event_mask_from_flat_batch;

pub use summary_events::summary_event_mask_from_tokens;

pub use summary_events::summary_event_mask_tensor;

pub use tokenizer::char_vocab::CharVocab;

Modules

api

Curated language-facing Dragon API.

bitnet_artifact

config

generation

inference

loss

summary_events

tokenizer

Structs

BDH

BDHConfig

ModelState

SequenceKernelConfig

Enums

SequenceMemorySystem

SequenceTrainingExecutor