MACE: Multicore Agentic Co-Design Engine
MACE User Manual
Introduction
Installation
Quick Start
Interactive Shell
Running the Loop
Baselines
Results and Metrics
LLM Backends
Code Coverage
Troubleshooting
MACE Requirements Specification
MACE Design Document
chia_openpiton Adapter
Supported Cores
MACE Evaluation
MACE: Multicore Agentic Co-Design Engine
MACE User Manual
View page source
MACE User Manual
Contents:
Introduction
Who this manual is for
How the parts fit together
Cores
Baselines
Where to start
Installation
Option 1: Nix flake
Option 2: conda and pip
Check the Python install
Clone OpenPiton
Patch the checkout
Quick Start
Set up the LLM backend
Run the loop
Flags
Read the output
Other cores and meshes
Interactive Shell
Commands
Credentials with mace init
Starting the shell
Shell commands
Core compatibility checks
Session defaults
Script mode
Running the Loop
MaceSpec
Gate workloads
Simulation limits
Budget
Planner directives
Driving the loop from Python
Baselines
Baseline (a): manual mesh scaling
Baseline (b): one-shot LLM
Running (b) and (c) back to back
Seeded-failure runs
Results and Metrics
Tables
Run statuses
The five metrics
mace results
SQL
LLM Backends
Backends
Selecting a backend and model
Vertex Gemini
Key-based backends
Cost reporting
Code Coverage
Requirement
Enabling coverage
Where coverage.dat lands
Reports
Coverage in the shell
Troubleshooting
Rebuilds and stale models
Ray times out during startup
Memory and MAKEFLAGS for 4x4 builds
Verilator versions
Tool name length
Other build errors