From af2ee7c7ce322c2f314e5a21be724a4c27863f67 Mon Sep 17 00:00:00 2001 From: Lars Bogner Date: Thu, 30 Jul 2026 16:40:02 +0200 Subject: [PATCH] Add gumbel router configs sweeping learnable-knob combinations Extends the 10-expert EnergyRouter + physical-conditioning benchmark config with the new opt-in gumbel combine weights, isolating the learn_centers/learn_temperature axis: none, centers only, and centers+temperature. Co-Authored-By: Claude Sonnet 5 --- ...er_energy_n10_gumbel_centers_physical.toml | 22 ++++++++++++++++++ ...ergy_n10_gumbel_centers_temp_physical.toml | 23 +++++++++++++++++++ .../router_energy_n10_gumbel_physical.toml | 22 ++++++++++++++++++ 3 files changed, 67 insertions(+) create mode 100644 configs/router_energy_n10_gumbel_centers_physical.toml create mode 100644 configs/router_energy_n10_gumbel_centers_temp_physical.toml create mode 100644 configs/router_energy_n10_gumbel_physical.toml diff --git a/configs/router_energy_n10_gumbel_centers_physical.toml b/configs/router_energy_n10_gumbel_centers_physical.toml new file mode 100644 index 0000000..386e01e --- /dev/null +++ b/configs/router_energy_n10_gumbel_centers_physical.toml @@ -0,0 +1,22 @@ +[train] +mode = "flow" +epochs = 30 +lr = 3e-4 +warmup_epochs = 3 +val_fraction = 0.1 +num_workers = 4 + +[model] +conditioning = "physical" +dropout = 0.0 + +[model.router] +enabled = true +type = "energy" +n_experts = 10 +expert_hidden_dim = 128 +expert_n_blocks = 4 +temperature = 0.05 +lambda_balance = 0.035 +learn_centers = true +gumbel = true diff --git a/configs/router_energy_n10_gumbel_centers_temp_physical.toml b/configs/router_energy_n10_gumbel_centers_temp_physical.toml new file mode 100644 index 0000000..933aa4b --- /dev/null +++ b/configs/router_energy_n10_gumbel_centers_temp_physical.toml @@ -0,0 +1,23 @@ +[train] +mode = "flow" +epochs = 30 +lr = 3e-4 +warmup_epochs = 3 +val_fraction = 0.1 +num_workers = 4 + +[model] +conditioning = "physical" +dropout = 0.0 + +[model.router] +enabled = true +type = "energy" +n_experts = 10 +expert_hidden_dim = 128 +expert_n_blocks = 4 +temperature = 0.05 +lambda_balance = 0.035 +learn_centers = true +learn_temperature = true +gumbel = true diff --git a/configs/router_energy_n10_gumbel_physical.toml b/configs/router_energy_n10_gumbel_physical.toml new file mode 100644 index 0000000..51c6a39 --- /dev/null +++ b/configs/router_energy_n10_gumbel_physical.toml @@ -0,0 +1,22 @@ +[train] +mode = "flow" +epochs = 30 +lr = 3e-4 +warmup_epochs = 3 +val_fraction = 0.1 +num_workers = 4 + +[model] +conditioning = "physical" +dropout = 0.0 + +[model.router] +enabled = true +type = "energy" +n_experts = 10 +expert_hidden_dim = 128 +expert_n_blocks = 4 +temperature = 0.05 +lambda_balance = 0.035 +learn_centers = false +gumbel = true