llmsim

Parallel discrete-event simulation for Python 3.14+.

llmsim pairs a fully typed, __slots__-based sequential simulation core with parallelism that comes from a share-nothing architecture — independent work across cores, never locks on the sequential engine.

Project status

llmsim is approaching its 1.0 API freeze. The sequential core, parallel replications, single-run PDES sharding, compute offload, and real-time mode are implemented and gated by same-seed determinism tests; the 1.0 tag and PyPI publish follow the Phase 5 example gallery. See the roadmap.

Install

llmsim is not on PyPI yet (the 1.0 publish is gated on Phase 5). Install the current source directly with pip:

pip install git+https://github.com/nobelk/llmsim

Once 1.0 is published, installation is simply pip install llmsim. Zero required runtime dependencies; Python 3.14+ on both GIL and free-threaded builds. See Getting started for your first model.

Where to start

• Getting started — install and run your first model in a dozen lines.
• Core concepts — the Sim, events, processes, and conditions.
• Which parallelism do I need? — the decision tree for choosing between replications, offload, and PDES.
• Migrating from SimPy 3 — the concept-by-concept porting guide, with all three canonical models worked side by side.

Reference and design

• API reference — every public symbol, generated from docstrings.
• Performance — measured scaling curves with their slowdown regimes.
• Deprecation policy — what the 1.x series promises.
• Design document — the full design everything derives from.
• SimPy 3 comparison — how llmsim relates to SimPy 3.