change to constant strategy

myolmoe/modeling_myolmoe.py

@@ -65,8 +65,8 @@ class OlmoeConfig(PretrainedConfig):
         small_expert_intermediate_ratio=64,
         small_expert_count=64,
         small_expert_sparsity_coef=0.1,
-        small_expert_strategy="increment",  # NEW
-        max_small_expert_count=64,         # NEW: total possible small experts
+        small_expert_strategy="constant",  # increment
+        max_small_expert_count=64,
         **kwargs,
     ):
         self.vocab_size = vocab_size