Add LP-relaxation and TRW-S approximate sharding solvers#484

Open

AlbedoWang wants to merge 29 commits into

mainfrom

kaijian/approx_solver_pr

Commits on May 26, 2026

Add Qwen3 AutoParallel model and examples
AlbedoWang
committed

Commits on May 28, 2026

Commits on May 29, 2026

Checkpoint scratch LP benchmark and ignore reference PDFs
AlbedoWang
committed

Commits on May 31, 2026

Merge remote-tracking branch 'origin/kaijian/annotated_search' into kaijian/merge_joint_opt
AlbedoWang
committed
Merge remote-tracking branch 'origin/kaijian/dp_solver' into kaijian/merge_joint_opt
AlbedoWang
committed
Integrate prune + dp_solver + annotated into a joint optimization
AlbedoWang
committed
Fix loaded-optimizer resolve() under dp_solver profiling
AlbedoWang
committed
Apply the memory constraint in get_lower_bound
AlbedoWang
committed
Add joint-optimization benchmark for LLaMA3 on 2D/3D meshes
AlbedoWang
committed
Extend benches for 3D: MODEL=small, MERGED flag, LP-bound certificate
AlbedoWang
committed
Use HiGHS (scipy.linprog) for the 3D LP-bound certificate
AlbedoWang
committed
Skip enumeration redistribute-cost computation (algorithm-preserving fast build)
AlbedoWang
committed
Store cluster_links node-level (drop per-option expansion) + DecisionVar slots
AlbedoWang
committed
Parallelize decision-var cost computation across forked workers
AlbedoWang
committed
Add cross-size prune+dp benchmark (latency + LP-relaxation accuracy)
AlbedoWang
committed

Commits on Jun 6, 2026

Add LP-relaxation solver, optimality-check gap logging, step-time profiling
AlbedoWang
committed

Commits on Jun 7, 2026

Commits on Jun 8, 2026