potential fix

Author: brb-nv

cpp/tensorrt_llm/executor/cache_transmission/cacheSplitConcat.cu

@@ -211,7 +211,9 @@ TargetRanksInfo TargetRanksInfoForDP(
         {
             for (int k = peerPPRankStart; k < peerPPRankEnd; k++)
             {
-                int irank = (k * peerTPNum * peerCPNum) + (j * peerTPNum) + i;
+                // Rank formula: ppRank * (tpNum * cpNum) + tpRank * cpNum + cpRank
+                // where i=tpRank, j=cpRank, k=ppRank
+                int irank = (k * peerTPNum * peerCPNum) + (i * peerCPNum) + j;
                 retRanks.push_back(irank);
             }
         }

tests/integration/defs/accuracy/test_disaggregated_serving.py

@@ -872,8 +872,9 @@ def test_auto_dtype(self, overlap_scheduler, mtp_nextn):
             task.evaluate(llm)
 
     @pytest.mark.skip_less_device(8)
-    @pytest.mark.parametrize("gen_pp,gen_tp,gen_cp", [(1, 2, 2), (2, 1, 2)],
-                             ids=["pp1tp2cp2", "pp2tp1cp2"])
+    @pytest.mark.parametrize("gen_pp,gen_tp,gen_cp", [(1, 1, 4), (1, 2, 2),
+                                                      (2, 1, 2)],
+                             ids=["pp1tp1cp4", "pp1tp2cp2", "pp2tp1cp2"])
     @pytest.mark.parametrize("cuda_graph_config", [
         None,
         {
@@ -911,6 +912,7 @@ def test_auto_dtype_with_helix(self, comms_medium, cuda_graph_config,
             "cache_transceiver_config": {
                 "backend": "UCX"
             },
+            # "print_iter_log": True,
         }
         gen_server_config = {
             "tensor_parallel_size": gen_tp,
@@ -929,6 +931,7 @@ def test_auto_dtype_with_helix(self, comms_medium, cuda_graph_config,
             "cache_transceiver_config": {
                 "backend": "UCX"
             },
+            # "print_iter_log": True,
         }
         disaggregated_server_config = {
             "hostname": "localhost",