started cleanup groupby

src/ops/group_by.cc

@@ -42,8 +42,6 @@ using Legion::TaskLauncher;
 using PCG::Node;
 
 
-// TODO replace "n" by "num_local_experts" everywhere in the script (only did it partially)
-
 void FFModel::group_by(const Tensor input,
                        const Tensor assign,
                        Tensor *outputs,
@@ -70,15 +68,13 @@ void FFModel::group_by(const Tensor input,
       dims[i] = input->dims[i];
     }
 
-//    // Define max expert capacity
+    // Define max expert capacity
     if (alpha != 0.0f) {
-//      int seq_len = input->dims[num_dims - 1];
       dims[num_dims - 1] = (int)ceil(alpha * k_experts_per_tok / num_local_experts * input->dims[num_dims - 1]);
 
     } else { // MK: added this for dummy groupby
       dims[num_dims - 1] = 128; // TODO remove magic number
     }
-    // printf("ff.groupby output: %d %d %d %d\n", dims[0], dims[1], dims[2], dims[3]);
 
     for (int i = 0; i < num_local_experts; i++) {
       // Creating one tensor per expert, each with size (DATA_DIMS,
@@ -163,22 +159,11 @@ Group_by::Group_by(FFModel &model,
     dims[i] = inputs[0]->dims[i];
   }
   // set max expert size
-  // TODO: this is a dirty fix while we don't use aggregate
 
-  // dims[num_dims - 2].size = (int)ceil(alpha * k_experts_per_tok / n * inputs[0]->dims[1].size);
-  // MK why is this - 2 instead of - 1? Also, why no alpha?
-
   if (alpha != 0.0f) {
-//      int seq_len = input->dims[num_dims - 1];
       dims[num_dims - 2].size = (int)ceil(alpha * k_experts_per_tok / n * inputs[0]->dims[2].size); // was inputs[0]->dims[1].size
 
-  } else { // MK: added this for dummy groupby
-    dims[num_dims - 2].size = 128;  // TODO remove magic number
-    //dims[num_dims - 2] = 128;
   }
-  // printf("groupby op output: %d %d %d %d\n", dims[0].size, dims[1].size, dims[2].size, dims[3].size);
-  // dims[num_dims - 2].size = 128;
-//  printf("max num tokens dim in output used to be %d\n", (int)ceil(alpha * k_experts_per_tok / n * inputs[0]->dims[1].size));
 
   for (int i = 0; i < n; i++) {
     outputs[i] = model.create_parallel_tensor_legion_ordering(