Papers › Three Phases of Expert Routing: How Load Balance Evolves During Mixture-of-Experts Training

Three Phases of Expert Routing: How Load Balance Evolves During Mixture-of-Experts Training

5 Apr 2026arXiv:2604.04230added by Syntology

Charafeddine Mouzouni

Title, abstract, authors and date from arXiv's metadata (CC0); this paper is not in the Papers with Code archive (frozen 2025-07-28).

We model Mixture-of-Experts (MoE) token routing as a congestion game with a single effective parameter, the congestion coefficient gamma_eff, that quantifies the balance-quality tradeoff. Tracking gamma_eff across training checkpoints of two open-source MoE models, OLMoE-1B-7B (20 checkpoints, with dense sampling in the surge region) and OpenMoE-8B (6 checkpoints), reveals a three-phase trajectory: a surge phase where the router learns to balance load (gamma_eff: 14 to 36-39, peaking in the step 30K-40K region), a stabilization phase where experts specialize under steady balance (B_0: 2.4 to 2.3, steps 100K-400K), and a relaxation phase where the router trades balance for quality as experts differentiate (gamma_eff: 27 to 9, steps 400K-1.2M). This non-monotone trajectory, invisible to post-hoc analysis of converged models, reveals that early MoE training prioritizes balance while late training prioritizes quality. The theoretical framework is honest about its limits: the single-type equilibrium reduces to temperature-scaled softmax (held-out L1: MFG = 0.199 vs. softmax = 0.200). The game is not a better predictor; it reveals what the temperature means and, critically, how that temperature evolves. We complement the dynamics with an effective congestion decomposition, a multi-type extension that improves load prediction via token clustering on all 16 layers (mean: 30%), scope diagnostics (K/M, epsilon_l), and robustness verification across four independent quality estimators (r >= 0.89). All confidence intervals are from bootstrap resampling over 50 independent text batches.

PaperPDFCode Syntology ran

In Syntology View this paper on Syntology: its repositories, every harvested function with whether it ran, its licence and the call to fetch it.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, Syntology's MCP tool lists every function and class Syntology harvested from this paper and whether it ran (how to connect): get_harvested_code_for_paper(arxiv_id="2604.04230")

Code

Syntology Ran 18 of 21 code samples harvested from 1 repository linked to this paper; 3 have no recorded run. Of those that ran: 18 ran with no contract checked.

By repository: found in paper text by Syntology: 21 samples from 1 repository, 18 ran. The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Cmouzouni/three-phases-moe found in paper text by Syntology report

Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

21 samples harvested; 18 ran; 0 honoured the contract we drafted; 3 have no recorded run. Read from Syntology's graph 2026-09-25; that is when this build read the record, not when the samples ran.

18ran
3unverified

Licence: 0 of the 21 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from Cmouzouni/three-phases-moe. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself; “our draft was wrong” and “fixture could not drive it” are failures of Syntology's instrument, not of the code.

Each sample ends with its code_sha256, Syntology's identity for that exact code. An agent fetches the stored sample with Syntology's MCP tool get_code(code_sha256="…") (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples from a repository marked official are listed first. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of four reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; or the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar). Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

compute_gamma_layer Cmouzouni/three-phases-moe/experiments/aws_quality_robustness.py found in paper text by Syntology ran MIT (permissive) · d8d4269c8c734c90 · report
compute_gamma_layer_split Cmouzouni/three-phases-moe/experiments/aws_quality_robustness.py found in paper text by Syntology ran MIT (permissive) · 84a88b95d8196d1f · report
entropy_norm Cmouzouni/three-phases-moe/experiments/exp3b_multi_model.py found in paper text by Syntology ran fingerprinted MIT (permissive) · ecf3322ebee8d2d2 · report
fit_gamma Cmouzouni/three-phases-moe/experiments/aws_olmoe_dense.py found in paper text by Syntology ran MIT (permissive) · b6fc027d85d02c48 · report
fit_gamma Cmouzouni/three-phases-moe/experiments/aws_openmoe_dynamics.py found in paper text by Syntology ran MIT (permissive) · d643137b9a8db451 · report
fit_gamma Cmouzouni/three-phases-moe/experiments/aws_quality_robustness.py found in paper text by Syntology ran MIT (permissive) · 1460c2c62a70e31e · report
fit_gamma_simple Cmouzouni/three-phases-moe/experiments/exp10b_cluster_baselines.py found in paper text by Syntology ran MIT (permissive) · 2f0aa4101587772e · report
fit_mixture_softmax Cmouzouni/three-phases-moe/experiments/exp10_definitive.py found in paper text by Syntology ran MIT (permissive) · dd49203038d1cfee · report
fit_temperature Cmouzouni/three-phases-moe/experiments/exp10_definitive.py found in paper text by Syntology ran fingerprinted MIT (permissive) · 98d36b41ed3f1a21 · report
fit_temperature Cmouzouni/three-phases-moe/experiments/exp10b_cluster_baselines.py found in paper text by Syntology ran fingerprinted MIT (permissive) · 62c1457f1d24c547 · report
maxvio Cmouzouni/three-phases-moe/experiments/exp3b_multi_model.py found in paper text by Syntology ran fingerprinted MIT (permissive) · 3ed98a39aaeae94e · report
mfg_social_cost Cmouzouni/three-phases-moe/moe/theory.py found in paper text by Syntology ran MIT (permissive) · 445f9db95ecf2b66 · report
per_cluster_softmax Cmouzouni/three-phases-moe/experiments/exp10b_cluster_baselines.py found in paper text by Syntology ran MIT (permissive) · 274d79df20322837 · report
show_c1_equivalence Cmouzouni/three-phases-moe/moe/theory.py found in paper text by Syntology ran MIT (permissive) · e6962f8185e939c0 · report
softmax Cmouzouni/three-phases-moe/moe/costs.py found in paper text by Syntology ran fingerprinted MIT (permissive) · 6a18510b8fe51466 · report
switch_aux_loss Cmouzouni/three-phases-moe/moe/theory.py found in paper text by Syntology ran MIT (permissive) · 9beb8cc0dfae2b64 · report
synthetic_quality Cmouzouni/three-phases-moe/moe/params.py found in paper text by Syntology ran MIT (permissive) · 377318b5516d076e · report
top_k_softmax Cmouzouni/three-phases-moe/moe/costs.py found in paper text by Syntology ran fingerprinted MIT (permissive) · 9d18b1ead51e244e · report
extract_and_analyze Cmouzouni/three-phases-moe/experiments/aws_openmoe_dynamics.py found in paper text by Syntology unverified MIT (permissive) · 642a49f96f69862d · report
extract_per_text_stats Cmouzouni/three-phases-moe/experiments/exp11b_dynamics_scaled.py found in paper text by Syntology unverified MIT (permissive) · f9db0ee626816365 · report
process_checkpoint Cmouzouni/three-phases-moe/experiments/aws_olmoe_dense.py found in paper text by Syntology unverified MIT (permissive) · 5b20d854dd88346d · report

Results from the paper

The Papers with Code archive ends with its 2025-07-28 snapshot. This paper's arXiv identifier, 2604.04230, was issued in April 2026, after that date, so the archive has no leaderboard rows for it.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections