Executor-Tyrant-Framework/Condensate

Revert PIPELINE to test_mode: true — SIGSEGV in production mode

Enable production condensation with recently_freed tombstone guard

Fix Cargo.toml default feature: preload-only build for LD_PRELOAD safety

Add system-wide laptop observer + keep-alive wrapper

Condensate: observer-data hardening — 4 membrane refinements (#260)

Fix condenser SIGSEGV — set test_mode=true in LD_PRELOAD pipeline

Fix LD_PRELOAD bootstrap SIGSEGV — static early-alloc buffer

Add *.so to .gitignore — compiled Rust binaries should not be tracked

Add GitHub Actions workflow: auto-sync to HF Space on push to main

Fix deposit(): use ng_tract.deposit_outcome, not write_outcome + Python I/O

Born dormant: grace period replaces skip list for system-wide safety

Skip Node.js processes in LD_PRELOAD hooks — fix gateway SEGV

Lazy neighbor cleanup: remove O(N) scan from every free

Exorcise field ghosts: remove_region on free, dead allocations leave the field

Fix LD_PRELOAD deadlock: static .bss ring buffer, zero heap allocation

Lock-free ring buffer for LD_PRELOAD hooks — eliminate malloc overhead

Condensate v2: Full Rust conversion — 12 modules, 105 tests, zero Python inflation

Wire Lenia field into pipeline — continuous thermal dynamics live

Lenia thermal field: continuous dynamics for memory temperature

Tuned pipeline: adaptive identity, predictions firing

Wire membrane hooks into full pipeline — the system is alive

Pipeline: the living loop — membrane → graph → predictor → condenser

Rust condenser: HOT/WARM/COLD tier management with LZ4 compression

Rust membrane: LD_PRELOAD system-level memory interception

Split load/train: CPU model load + GPU training separately

Rust-Python bridge: maturin build, wrapper, head-to-head benchmark

Rust core: AccessGraph + RustPredictor with PyO3 bindings

Fix: increase GPU timeout to 180s, remove torch_dtype kwarg

Restructure: single GPU call for load+train, no nesting

Fix ZeroGPU: load model on CPU, move to GPU inside decorated fns

Head-level membrane v2 — decompose attention by individual heads

Upgrade to gpt2-large (774M, 36 layers, 20 heads)

Upgrade to gpt2-medium (355M params, 24 layers)

Remove show_copy_button — not in Gradio 6

Restore full PyTorch demo with ZeroGPU support

Remove PyTorch dependency — demo uses PoC layers directly

Use Gradio 6.10.0, explicit server bind

Fix Gradio version: use sdk_version 5.29.0, don't pin gradio in requirements

Add .gitignore, include tests, remove cache artifacts

Fix HF Space: CPU-only torch, lazy imports

Condensate PoC — all 4 layers + HF Spaces demo

initial commit