How sivabalajisleo/leo works
HYPER is a research and execution system for finding lower-cost software pathways that satisfy an application contract on constrained local hardware; it does not emulate GPU hardware. The principal workflow accepts a workload and contract, analyzes necessary work, searches candidate pathways, executes and verifies a candidate, falls back when needed, and records evidence and measurements. The README also describes prediction, precision, reuse, and heterogeneous scheduling as capabilities. This map reflects the documented workflow and sampled orchestration; several subsystems are represented at their exact indexed paths because their implementations were not excerpted.
Main entry points: pipeline.py HYPER-X pipeline · contract.py Contract IR · master_pipeline.py Research pipeline · mvp_pipeline.py Adaptive runtime
Components 25
Contract workflow
- HYPER-X pipeline
hyper_x/pipeline.py - Contract IR
hyper_x/contract_ir/contract.py
Work reduction
- Information boundary
hyper_x/information_boundary/engine.py - Necessary work
hyper_x/necessity/compiler.py - Pathway search
hyper_x/pathway_search/search_engine.py - Exact reuse
hyper_x/pathway_search/exact_reuse.py - Representation search
hyper_x/representations/representation_search.py - Algorithm discovery
hyper_x/discovery/discovery_engine.py - Prediction
hyper_x/prediction/prediction_engine.py - Reconstruction
hyper_x/reconstruction/reconstruction_engine.py - Computational escape
hyper/escape/escape_engine.py - Math adapter
hyper_x/ahce/domains/math.py - Residual computation
hyper_cco/residual_engine.py
Verification and evidence
- Candidate verifier
hyper_x/verification/verifier.py - Fallback execution
hyper_x/fallback/engine.py - Execution certificates
hyper/certificates/certificate_engine.py - Research pipeline
hyper/research_engine/master_pipeline.py - Anti-cheat audit
hyper/research_engine/anti_cheat_and_holdout.py
Runtime and measurement
- Work avoidance tracking
backend/analytics/avoidance_tracker.py - Local CPU and iGPU
- Runtime telemetry
hyper_runtime/benchmarking/telemetry_tracker.py - Adaptive runtime
hyper_runtime/mvp_pipeline.py - Semantic replay cache
hyper_runtime/semantic_replay/replay_cache.py - CPU kernel scheduler
hyper_runtime/cpu_orchestrator/kernel_scheduler.py
Other components
- Workload user
Connections 30
- Workload user → HYPER-X pipeline: submits workload
README.md - HYPER-X pipeline → Contract IR: parses contract
hyper_x/pipeline.py - HYPER-X pipeline → Information boundary: analyzes workload
hyper_x/pipeline.py - HYPER-X pipeline → Necessary work: builds work DAG
hyper_x/pipeline.py - HYPER-X pipeline → Representation search: evaluates representations
hyper_x/pipeline.py - HYPER-X pipeline → Pathway search: searches candidates
hyper_x/pipeline.py - Pathway search → Exact reuse: checks exact reuse
hyper_x/pathway_search/search_engine.py - Pathway search → Contract IR: constrains candidates
hyper_x/pathway_search/search_engine.py - HYPER-X pipeline → Algorithm discovery: uses discovery
hyper_x/pipeline.py - HYPER-X pipeline → Prediction: uses prediction
hyper_x/pipeline.py - HYPER-X pipeline → Reconstruction: uses reconstruction
hyper_x/pipeline.py - HYPER-X pipeline → Candidate verifier: verifies candidate
README.md - HYPER-X pipeline → Fallback execution: runs fallback
hyper_x/pipeline.py - HYPER-X pipeline → Local CPU and iGPU: schedules execution
- HYPER-X pipeline → Execution certificates: records proof
README.md - HYPER-X pipeline → Work avoidance tracking: tracks avoided work
- HYPER-X pipeline → Runtime telemetry: records measurements
README.md - Research pipeline → Contract IR: loads workload contract
hyper/research_engine/master_pipeline.py - Research pipeline → Algorithm discovery: uses discovery graph
hyper/research_engine/master_pipeline.py - Research pipeline → Anti-cheat audit: audits claims
hyper/research_engine/master_pipeline.py - Research pipeline → Local CPU and iGPU: runs reference workload
hyper/research_engine/master_pipeline.py - Adaptive runtime → Semantic replay cache: checks replay
hyper_runtime/mvp_pipeline.py - Adaptive runtime → CPU kernel scheduler: dispatches kernels
hyper_runtime/mvp_pipeline.py - Adaptive runtime → Runtime telemetry: records runtime metrics
hyper_runtime/mvp_pipeline.py - Adaptive runtime → Local CPU and iGPU: executes workload
hyper_runtime/mvp_pipeline.py - Math adapter → Local CPU and iGPU: executes matrix product
hyper_x/ahce/domains/math.py - Math adapter → Computational escape: uses low-rank method
- Computational escape → Contract IR: checks constraints
hyper/escape/escape_engine.py - Residual computation → Contract IR: uses residual contract
hyper_cco/residual_engine.py - Candidate verifier → Fallback execution: rejects to fallback
hyper_x/pipeline.py