PHASE3.md

Phase 3 — CPU Runtime (Complete in-tree)

Status: Implemented in repository (verify locally via README build steps)

Delivered

Area	Implementation
Memory	uaii-memory: budget, arena, tensor pool, allocator
Scheduler	CpuScheduler — CPU placement, plan order
Execution engine	runtime::Session lifecycle (create / feed / run / fetch)
CPU backend	backends::CpuBackend
Kernels	MatMul, Softmax, LayerNorm, RMSNorm, Relu, Gelu, Silu, Add, Mul, Identity
Transformer-style	uaii run --demo tiny_block (LN → MatMul → Softmax → MatMul → Add → RMSNorm)
Toy E2E	uaii run --demo toy_mlp with deterministic expected output

Verify locally

cmake -S . -B build -DUAII_BUILD_TESTS=ON -DUAII_BUILD_PLUGINS=ON
cmake --build build --config Release --parallel

uaii run --demo toy_mlp
uaii run --demo tiny_block

uaii run examples/ir/toy_mlp.uaii.json \
  --weight-init ones \
  --input x=1,2,3,4 \
  --output y_prob

Expected for toy_mlp demo: y_prob = [0.25, 0.25, 0.25, 0.25].