Ravana
Modules
A map of the three source packages. Paths are relative to each package's src/ directory. Only the most load-bearing modules are listed; the full tree is discove
Modules
A map of the three source packages. Paths are relative to each package's src/
directory. Only the most load-bearing modules are listed; the full tree is
discoverable under ravana/src, ravana_ml/src, ravana-v2/src.
ravana_ml — ML substrate (ravana_ml/src/ravana_ml/)
| Module | Responsibility |
|---|---|
tensor.py | CPU-native autodiff tensor (the torch-like drop-in). |
nn/ | module.py, functional.py, neural_decoder.py (the chat decoder), rlm.py / rlm_v2*.py (graph relation learner + verb-offset, entity adapters, sleep). |
graph.py | ConceptGraph: typed nodes/edges, find_similar, spread_activation, pruning. |
embedder.py | GloVe→64-D projection, word embeddings. |
ontology.py | ConceptNet + attribute/Binder encoder, Lancaster probe. |
free_energy.py, propagation.py, plasticity.py | Free-energy learning core. |
currencies.py / currency.py | Value/neuromodulator bookkeeping. |
episode_injector.py | Episodic memory injection. |
ravana — chat engine (ravana/src/ravana/)
| Subpackage | Key modules |
|---|---|
chat/ | engine.py (CognitiveChatEngine — orchestrator that composes 8 mixins), engine_generation.py (GenerationMixin), engine_graph.py (GraphMixin), engine_memory.py (MemoryMixin), engine_monitor.py (MonitorMixin), engine_persistence.py (PersistenceMixin), engine_reasoning.py (ReasoningMixin), engine_self_query.py (SelfQueryMixin), engine_web_search.py (WebSearchMixin). Plus brain_regions.py (brain-repair prepasses), consistency_monitor.py (cross-turn self-consistency check), harm_intent_gate.py (pre-generation safety classifier), support_router.py (advice/wellbeing routing), personal_fact_store.py (learned user-profile facts), user_model.py (opinions/corrections/stance), interface.py (interactive chat entrypoint), coherence_gate.py, junk_scorer.py, monitor_gate.py, intent_router.py, chain_walker.py, response_gen.py, self_model_router.py, salad_classifier.py, web_learning.py, constants.py. |
core/ | emotion.py (VAD), identity.py, meaning.py, sleep.py, predictive_coding.py, hrr_reasoner.py, analogy_engine.py, causal_schema.py, working_memory.py, dual_process.py, global_workspace.py, meta_cognition.py, mirror.py (emotional mirror / theory-of-mind). Reasoning/memory additions: fact_reasoning.py (chain-walking + conditionals over episodic memory), frequency_model.py (learned word-frequency model, replaces hand word lists), in_prompt_reasoner.py, multi_hop_reasoner.py, temporal_cloze.py (TimeDial-style blank filling), temporal_grounding.py, temporal_reasoner.py (event ordering / time-logic), situation_model.py, event_schema.py, proposition_parser.py, question_decomposition.py, sub_answer_synthesizer.py, hippocampal_buffer.py (episodic buffer + FactTriple), relation_memory.py, attractor_memory.py, dual_code_space.py, active_inference.py, abstraction_engine.py, implicature_detector.py, quantity_modifier.py, coherence.py, vsa.py, system1.py, system2.py, belief_store.py. |
language/ | surface_realizer.py, syntactic_cell_assembly.py, basal_ganglia.py (gate), cerebellar_ngram.py, prefrontal_workspace.py, verb_lexicon.py, register.py, schemas.py. |
cognitive/ | Re-exports the GRACE cognitive core from ravana_grace.core. |
ontology/ | derived.py, graph_typing.py, conceptnet.py, linggen.py (LingGen P6 sensorimotor conditioning + LingGenConditioner). |
web/ | web_to_graph.py, learner.py, openie.py (web → typed edges). |
decoder/ | engine.py, predictive_coding_generator.py (settle path). |
graph/ | engine.py (GraphEngine wrapper). |
bootstrap/ | manager.py, pmi_seeder.py (cold-start seeding). |
learn/ | consolidation.py, curiosity.py. |
world/ | world-model state. |
ravana_grace — GRACE governor (ravana-v2/src/ravana_grace/)
| Subpackage | Key modules |
|---|---|
core/ | governor.py (orchestrator), identity.py, emotion.py, meaning.py, sleep.py, global_workspace.py, belief_reasoner.py, active_epistemology.py, meta_cognition.py, strategy.py, strategy_learning.py, planning.py, environment.py, predictive_world.py, reality_friction.py, social_epistemology.py, surgical_probes.py, vector_index.py, dual_process.py, empathy.py, memory.py, human_memory.py, memory_reconstructor.py, resolution.py, occam_layer.py, hypothesis_generation.py, conversational_repair.py, dialogue_context.py, state.py, runtime.py. |
dialogue/ | dialogue_engine.py. |
agent/ | ravana_agent_loop.py, mode_orchestrator.py, version_manager.py, test_runner.py. |
interface_agent/ | scripts/ — LLM interpreter, memory learner, reality grounding, telegram reporter, prompt composer (optional LLM bridges). |
probes/ | constraint_stress.py, exploration_pressure.py, learning_signal.py. |
research/ | core_k0/ — agent-loop research harness (k0…k3 experiments). |
training/ | training loops for the governor. |
Scripts (scripts/)
See BENCHMARKS.md for the full list and how to run each.
Experiments (experiments/)
experiments/ is a package of research harnesses (experiment_cross_domain.py,
benchmark_arc.py, experiment_ablation.py, experiment_sleep_memory.py, …)
used by the benchmark/diagnostic scripts. They are not part of the runtime path
but are the reference implementations behind the numbers in
BENCHMARKS.md.
Theoretical Foundations
RAVANA replaces the standard ML paradigm (loss functions → gradients → weight update) with a pressure-driven self-organization paradigm. This document explains
Training
All training lives in scripts/train.py (one file, four modes). The old iterative_train.py, train_decoder_phase2.py, and _linggen_train_big.py scripts were merge