diff --git a/docs/guides/vlm/nemotron-omni.md b/docs/guides/vlm/nemotron-omni.md
index 2ca682c306..eccd3c83dd 100644
--- a/docs/guides/vlm/nemotron-omni.md
+++ b/docs/guides/vlm/nemotron-omni.md
@@ -140,7 +140,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/audio_finetune/qwen3_omni_asr/ami_sft.yaml b/examples/audio_finetune/qwen3_omni_asr/ami_sft.yaml
index 91225e90a6..09de0ca3b4 100644
--- a/examples/audio_finetune/qwen3_omni_asr/ami_sft.yaml
+++ b/examples/audio_finetune/qwen3_omni_asr/ami_sft.yaml
@@ -66,8 +66,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    experts: te
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/audio_finetune/qwen3_omni_asr/multi_en_sft.yaml b/examples/audio_finetune/qwen3_omni_asr/multi_en_sft.yaml
index bad1787bd2..c62f9216a1 100644
--- a/examples/audio_finetune/qwen3_omni_asr/multi_en_sft.yaml
+++ b/examples/audio_finetune/qwen3_omni_asr/multi_en_sft.yaml
@@ -55,8 +55,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    experts: te
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/llm_benchmark/glm/glm_4.7_te_deepep.yaml b/examples/llm_benchmark/glm/glm_4.7_te_deepep.yaml
index f5ac8b1912..a3bf1fe426 100644
--- a/examples/llm_benchmark/glm/glm_4.7_te_deepep.yaml
+++ b/examples/llm_benchmark/glm/glm_4.7_te_deepep.yaml
@@ -79,7 +79,7 @@ model:
     rms_norm: te
     rope_fusion: true
     experts: gmm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: true
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_benchmark/qwen/qwen3_next_te_deepep.yaml b/examples/llm_benchmark/qwen/qwen3_next_te_deepep.yaml
index 2decc69ae4..f680a0ee64 100644
--- a/examples/llm_benchmark/qwen/qwen3_next_te_deepep.yaml
+++ b/examples/llm_benchmark/qwen/qwen3_next_te_deepep.yaml
@@ -78,7 +78,7 @@ model:
     rms_norm: te
     rope_fusion: true
     experts: gmm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: true
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml b/examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml
index 1368c7a7ab..124a8a831c 100644
--- a/examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml
+++ b/examples/llm_finetune/deepseek_v32/deepseek_v32_hellaswag_pp.yaml
@@ -73,7 +73,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag.yaml b/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag.yaml
index c52655bdbb..b8af2bba7e 100644
--- a/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag.yaml
+++ b/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag.yaml
@@ -76,7 +76,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: torch_mm
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag_lora.yaml b/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag_lora.yaml
index 7ebb18a80a..c5c1ce9758 100644
--- a/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag_lora.yaml
+++ b/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_hellaswag_lora.yaml
@@ -63,7 +63,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: torch_mm
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_packed_sequence_hellaswag.yaml b/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_packed_sequence_hellaswag.yaml
index 7454d361c4..ad518eb6c2 100644
--- a/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_packed_sequence_hellaswag.yaml
+++ b/examples/llm_finetune/deepseek_v4/deepseek_v4_flash_packed_sequence_hellaswag.yaml
@@ -71,7 +71,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: torch_mm
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_hellaswag_all_tilelang_pp8_ep64_20steps.yaml b/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_hellaswag_all_tilelang_pp8_ep64_20steps.yaml
index b6a5611fbd..f9fc1501b1 100644
--- a/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_hellaswag_all_tilelang_pp8_ep64_20steps.yaml
+++ b/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_hellaswag_all_tilelang_pp8_ep64_20steps.yaml
@@ -73,7 +73,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: torch_mm
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_packed_sequence_hellaswag_all_tilelang_pp8_ep64_1024_20steps.yaml b/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_packed_sequence_hellaswag_all_tilelang_pp8_ep64_1024_20steps.yaml
index 5f69c5f734..526690bc18 100644
--- a/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_packed_sequence_hellaswag_all_tilelang_pp8_ep64_1024_20steps.yaml
+++ b/examples/llm_finetune/deepseek_v4/deepseek_v4_pro_packed_sequence_hellaswag_all_tilelang_pp8_ep64_1024_20steps.yaml
@@ -73,7 +73,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: torch_mm
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/glm/glm_4.7_te_deepep.yaml b/examples/llm_finetune/glm/glm_4.7_te_deepep.yaml
index fc0e89881d..bcd5a113d1 100644
--- a/examples/llm_finetune/glm/glm_4.7_te_deepep.yaml
+++ b/examples/llm_finetune/glm/glm_4.7_te_deepep.yaml
@@ -37,6 +37,9 @@ rng:
 model:
   _target_: nemo_automodel.NeMoAutoModelForCausalLM.from_pretrained
   pretrained_model_name_or_path: zai-org/GLM-4.7
+  backend:
+    _target_: nemo_automodel.components.models.common.BackendConfig
+    dispatcher: hybridep
 
 checkpoint:
   enabled: false
diff --git a/examples/llm_finetune/glm/glm_5.1_hellaswag_pp.yaml b/examples/llm_finetune/glm/glm_5.1_hellaswag_pp.yaml
index 1a8c61f857..62c24899a2 100644
--- a/examples/llm_finetune/glm/glm_5.1_hellaswag_pp.yaml
+++ b/examples/llm_finetune/glm/glm_5.1_hellaswag_pp.yaml
@@ -49,7 +49,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/glm/glm_5.1_lora.yaml b/examples/llm_finetune/glm/glm_5.1_lora.yaml
index a20bce7138..065a938c83 100644
--- a/examples/llm_finetune/glm/glm_5.1_lora.yaml
+++ b/examples/llm_finetune/glm/glm_5.1_lora.yaml
@@ -28,7 +28,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: gmm
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
diff --git a/examples/llm_finetune/glm/glm_5_hellaswag_pp.yaml b/examples/llm_finetune/glm/glm_5_hellaswag_pp.yaml
index 27ab4e9c3e..5f03d28123 100644
--- a/examples/llm_finetune/glm/glm_5_hellaswag_pp.yaml
+++ b/examples/llm_finetune/glm/glm_5_hellaswag_pp.yaml
@@ -62,7 +62,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft.yaml b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft.yaml
index c0fb10ab78..5bc318a8a7 100644
--- a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft.yaml
+++ b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft.yaml
@@ -15,7 +15,8 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: 8
diff --git a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft_chat.yaml b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft_chat.yaml
index 0515ba5379..010ea363ce 100644
--- a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft_chat.yaml
+++ b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_full_sft_chat.yaml
@@ -15,7 +15,8 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: 8
diff --git a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft.yaml b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft.yaml
index 15ff30d4ab..2820598193 100644
--- a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft.yaml
+++ b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft.yaml
@@ -15,7 +15,7 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    dispatcher: torch
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: 1
diff --git a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft_packing.yaml b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft_packing.yaml
index 6795d9248f..a3f3d54354 100644
--- a/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft_packing.yaml
+++ b/examples/llm_finetune/gpt_oss/customizer_gpt_oss_peft_packing.yaml
@@ -15,7 +15,7 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    dispatcher: torch
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: 1
diff --git a/examples/llm_finetune/hy_v3/hy3_preview_deepep_lora.yaml b/examples/llm_finetune/hy_v3/hy3_preview_deepep_lora.yaml
index ad25bf6623..21607b4ead 100644
--- a/examples/llm_finetune/hy_v3/hy3_preview_deepep_lora.yaml
+++ b/examples/llm_finetune/hy_v3/hy3_preview_deepep_lora.yaml
@@ -46,7 +46,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     experts: torch_mm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: false
     gate_precision: float32
     enable_hf_state_dict_adapter: true
diff --git a/examples/llm_finetune/ling/ling_flash_2_0_sft.yaml b/examples/llm_finetune/ling/ling_flash_2_0_sft.yaml
index 326336d580..3e5928bc13 100644
--- a/examples/llm_finetune/ling/ling_flash_2_0_sft.yaml
+++ b/examples/llm_finetune/ling/ling_flash_2_0_sft.yaml
@@ -42,7 +42,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     experts: torch_mm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/mimo_v2_flash/mimo_v2_flash_hellaswag.yaml b/examples/llm_finetune/mimo_v2_flash/mimo_v2_flash_hellaswag.yaml
index 4e5d3a6941..9d4c62c7df 100644
--- a/examples/llm_finetune/mimo_v2_flash/mimo_v2_flash_hellaswag.yaml
+++ b/examples/llm_finetune/mimo_v2_flash/mimo_v2_flash_hellaswag.yaml
@@ -68,7 +68,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: torch_mm
     gate_precision: float32
     enable_hf_state_dict_adapter: true
diff --git a/examples/llm_finetune/minimax_m2/minimax_m2.1_hellaswag_pp.yaml b/examples/llm_finetune/minimax_m2/minimax_m2.1_hellaswag_pp.yaml
index 623c79e333..556aa2461e 100644
--- a/examples/llm_finetune/minimax_m2/minimax_m2.1_hellaswag_pp.yaml
+++ b/examples/llm_finetune/minimax_m2/minimax_m2.1_hellaswag_pp.yaml
@@ -63,7 +63,7 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: true
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: gmm
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
diff --git a/examples/llm_finetune/minimax_m2/minimax_m2.5_hellaswag_pp.yaml b/examples/llm_finetune/minimax_m2/minimax_m2.5_hellaswag_pp.yaml
index cb295b885c..7f4c8e604e 100644
--- a/examples/llm_finetune/minimax_m2/minimax_m2.5_hellaswag_pp.yaml
+++ b/examples/llm_finetune/minimax_m2/minimax_m2.5_hellaswag_pp.yaml
@@ -66,7 +66,7 @@ model:
     rms_norm: torch_fp32
     rope_fusion: true
     experts: torch_mm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_lora.yaml b/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_lora.yaml
index 621b7e1260..c0b0115b6d 100644
--- a/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_lora.yaml
+++ b/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_lora.yaml
@@ -53,7 +53,7 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: true
-    dispatcher: deepep
+    dispatcher: hybridep
     experts: gmm
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
diff --git a/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_pp.yaml b/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_pp.yaml
index 6a2d8eea54..f675b03c22 100644
--- a/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_pp.yaml
+++ b/examples/llm_finetune/minimax_m2/minimax_m2.7_hellaswag_pp.yaml
@@ -66,7 +66,7 @@ model:
     rms_norm: torch_fp32
     rope_fusion: false
     experts: torch_mm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft.yaml b/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft.yaml
index 78aa42b59a..fcd6bc1a20 100644
--- a/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft.yaml
+++ b/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft.yaml
@@ -15,7 +15,8 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: 8
diff --git a/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft_chat.yaml b/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft_chat.yaml
index 0ef677a9b9..7392246569 100644
--- a/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft_chat.yaml
+++ b/examples/llm_finetune/nemotron/customizer_nemotron_nano_full_sft_chat.yaml
@@ -15,7 +15,8 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: 8
diff --git a/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft.yaml b/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft.yaml
index 78cb713402..035b449953 100644
--- a/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft.yaml
+++ b/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft.yaml
@@ -15,7 +15,8 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: null
diff --git a/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft_packing.yaml b/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft_packing.yaml
index ce894a09e7..7ab97b3c73 100644
--- a/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft_packing.yaml
+++ b/examples/llm_finetune/nemotron/customizer_nemotron_nano_peft_packing.yaml
@@ -15,7 +15,8 @@ model:
   attn_implementation: sdpa
   backend:
     _target_: nemo_automodel.components.models.common.utils.BackendConfig
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager
   dp_size: null
diff --git a/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft.yaml b/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft.yaml
index e676c53dcc..8f8f0a16f6 100644
--- a/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft.yaml
+++ b/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft.yaml
@@ -63,7 +63,7 @@ model:
     linear: te
     rms_norm: te
     experts: gmm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_fsdp_optimizations: false
 
diff --git a/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft_gb200.yaml b/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft_gb200.yaml
index 499dace967..0821e87ecf 100644
--- a/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft_gb200.yaml
+++ b/examples/llm_finetune/nemotron/nemotron_ultra_v3_hellaswag_peft_gb200.yaml
@@ -133,3 +133,8 @@ lr_scheduler:
   lr_decay_style: cosine
   lr_warmup_steps: 10
   min_lr: 1.0e-5
+
+ci:
+  # pp_size(1) * ep_size(16) = 16 GPUs => 4 nodes (4 GB200/node).
+  recipe_owner: adil-a
+  nodes: 4
diff --git a/examples/llm_finetune/nemotron/nemotron_ultra_v3_squad.yaml b/examples/llm_finetune/nemotron/nemotron_ultra_v3_squad.yaml
index a0fa50eaae..c00b0a9d99 100644
--- a/examples/llm_finetune/nemotron/nemotron_ultra_v3_squad.yaml
+++ b/examples/llm_finetune/nemotron/nemotron_ultra_v3_squad.yaml
@@ -121,6 +121,11 @@ lr_scheduler:
   lr_warmup_steps: 10
   min_lr: 1.0e-6
 
+ci:
+  # pp_size(1) * ep_size(64) = 64 GPUs => 8 nodes (8 H100/node).
+  recipe_owner: adil-a
+  nodes: 8
+
 # wandb:
 #   project: my-project
 #   entity: my-entity
diff --git a/examples/llm_finetune/qwen/qwen3_next_te_deepep.yaml b/examples/llm_finetune/qwen/qwen3_next_te_deepep.yaml
index a53aafc78d..42f64ad56b 100644
--- a/examples/llm_finetune/qwen/qwen3_next_te_deepep.yaml
+++ b/examples/llm_finetune/qwen/qwen3_next_te_deepep.yaml
@@ -38,6 +38,9 @@ rng:
 model:
   _target_: nemo_automodel.NeMoAutoModelForCausalLM.from_pretrained
   pretrained_model_name_or_path: Qwen/Qwen3-Next-80B-A3B-Instruct
+  backend:
+    _target_: nemo_automodel.components.models.common.BackendConfig
+    dispatcher: hybridep
 
 checkpoint:
   enabled: false
diff --git a/examples/llm_finetune/stepfun/step_3.5_flash_hellaswag_pp.yaml b/examples/llm_finetune/stepfun/step_3.5_flash_hellaswag_pp.yaml
index 7192942154..de949b49d7 100644
--- a/examples/llm_finetune/stepfun/step_3.5_flash_hellaswag_pp.yaml
+++ b/examples/llm_finetune/stepfun/step_3.5_flash_hellaswag_pp.yaml
@@ -65,7 +65,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: true
-    enable_deepep: true
+    experts: gmm
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/llm_pretrain/deepseekv3_pretrain.yaml b/examples/llm_pretrain/deepseekv3_pretrain.yaml
index 6dae590ff5..bff57a442b 100644
--- a/examples/llm_pretrain/deepseekv3_pretrain.yaml
+++ b/examples/llm_pretrain/deepseekv3_pretrain.yaml
@@ -152,3 +152,8 @@ lr_scheduler:
   lr_decay_style: cosine
   lr_warmup_steps: 500
   min_lr: 0.0
+
+ci:
+  # pp_size(8) * ep_size(16) = 128 GPUs => 16 nodes (8 H100/node).
+  recipe_owner: hemildesai
+  nodes: 16
diff --git a/examples/vlm_finetune/kimi/kimi25vl_medpix.yaml b/examples/vlm_finetune/kimi/kimi25vl_medpix.yaml
index 3c9257b7c1..f3f63f5cbe 100644
--- a/examples/vlm_finetune/kimi/kimi25vl_medpix.yaml
+++ b/examples/vlm_finetune/kimi/kimi25vl_medpix.yaml
@@ -62,7 +62,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml b/examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml
index 5fc11c377a..1abcdef428 100644
--- a/examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml
+++ b/examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml
@@ -61,7 +61,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: true
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/vlm_finetune/qwen3/qwen3_omni_moe_30b_te_deepep.yaml b/examples/vlm_finetune/qwen3/qwen3_omni_moe_30b_te_deepep.yaml
index 8ad20de42f..11486c4f9b 100644
--- a/examples/vlm_finetune/qwen3/qwen3_omni_moe_30b_te_deepep.yaml
+++ b/examples/vlm_finetune/qwen3/qwen3_omni_moe_30b_te_deepep.yaml
@@ -45,8 +45,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    experts: te
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/qwen3/qwen3_vl_moe_235b.yaml b/examples/vlm_finetune/qwen3/qwen3_vl_moe_235b.yaml
index fbaa242b24..ffb9602e56 100644
--- a/examples/vlm_finetune/qwen3/qwen3_vl_moe_235b.yaml
+++ b/examples/vlm_finetune/qwen3/qwen3_vl_moe_235b.yaml
@@ -46,7 +46,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_neat_packing.yaml b/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_neat_packing.yaml
index aabc91bd17..3e91749092 100644
--- a/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_neat_packing.yaml
+++ b/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_neat_packing.yaml
@@ -44,8 +44,8 @@ model:
     linear: te
     rms_norm: te
     rope_fusion: false
-    experts: te
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_te_deepep.yaml b/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_te_deepep.yaml
index ee2f7bff9e..2a6d1db089 100644
--- a/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_te_deepep.yaml
+++ b/examples/vlm_finetune/qwen3/qwen3_vl_moe_30b_te_deepep.yaml
@@ -47,8 +47,8 @@ model:
     linear: te
     rms_norm: torch_fp32
     rope_fusion: false
-    experts: te
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b.yaml b/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b.yaml
index bc0006a2ac..59c43f154f 100644
--- a/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b.yaml
+++ b/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b.yaml
@@ -44,7 +44,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b_neat_packing.yaml b/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b_neat_packing.yaml
index 7c9c326f77..cc19c1e258 100644
--- a/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b_neat_packing.yaml
+++ b/examples/vlm_finetune/qwen3_5_moe/qwen3_5_35b_neat_packing.yaml
@@ -49,7 +49,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/qwen3_5_moe/qwen3_5_moe_medpix.yaml b/examples/vlm_finetune/qwen3_5_moe/qwen3_5_moe_medpix.yaml
index 275bd0ec5d..cc7dd9fe26 100644
--- a/examples/vlm_finetune/qwen3_5_moe/qwen3_5_moe_medpix.yaml
+++ b/examples/vlm_finetune/qwen3_5_moe/qwen3_5_moe_medpix.yaml
@@ -45,7 +45,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: false
+    experts: gmm
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/qwen3_5_moe/qwen3_6_35b.yaml b/examples/vlm_finetune/qwen3_5_moe/qwen3_6_35b.yaml
index dc5a3d9066..9455cda80a 100644
--- a/examples/vlm_finetune/qwen3_5_moe/qwen3_6_35b.yaml
+++ b/examples/vlm_finetune/qwen3_5_moe/qwen3_6_35b.yaml
@@ -48,7 +48,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: true
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/examples/vlm_finetune/stepfun/step3p7_medpix_200b_ep32pp4.yaml b/examples/vlm_finetune/stepfun/step3p7_medpix_200b_ep32pp4.yaml
index eb913a679a..020dd7613c 100644
--- a/examples/vlm_finetune/stepfun/step3p7_medpix_200b_ep32pp4.yaml
+++ b/examples/vlm_finetune/stepfun/step3p7_medpix_200b_ep32pp4.yaml
@@ -45,7 +45,7 @@ model:
     rms_norm: torch_fp32
     rope_fusion: false
     experts: gmm
-    dispatcher: deepep
+    dispatcher: hybridep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
     enable_fsdp_optimizations: true
diff --git a/fern/versions/nightly/pages/guides/vlm/nemotron-omni.mdx b/fern/versions/nightly/pages/guides/vlm/nemotron-omni.mdx
index 30ec1e7644..704f847b39 100644
--- a/fern/versions/nightly/pages/guides/vlm/nemotron-omni.mdx
+++ b/fern/versions/nightly/pages/guides/vlm/nemotron-omni.mdx
@@ -144,7 +144,8 @@ model:
     linear: torch
     rms_norm: torch_fp32
     rope_fusion: false
-    enable_deepep: false
+    experts: gmm
+    dispatcher: deepep
     fake_balanced_gate: false
     enable_hf_state_dict_adapter: true
 
diff --git a/nemo_automodel/components/models/common/utils.py b/nemo_automodel/components/models/common/utils.py
index 72d51a4fb4..e3ba879056 100644
--- a/nemo_automodel/components/models/common/utils.py
+++ b/nemo_automodel/components/models/common/utils.py
@@ -14,7 +14,6 @@
 
 import importlib.util
 import logging
-import warnings
 from contextlib import contextmanager, nullcontext
 from dataclasses import dataclass
 from typing import Any, Literal
@@ -176,7 +175,8 @@ class BackendConfig:
             manager instance across MoE layers.
         dispatcher_async_dispatch: Whether DeepEP/UCCL-EP dispatch should return asynchronously
             and allocate dispatched tensors on the communication stream.
-        enable_deepep: Deprecated. Use dispatcher="deepep" and experts="gmm" instead.
+        enable_deepep: Removed and ignored. Logs a warning if set; configure "dispatcher"
+            and "experts" explicitly instead.
         fake_balanced_gate: If True, replace the learned Gate with FakeBalancedGate
             that assigns tokens to experts without learned routing weights.
         fake_gate_noise: Noise level [0, 1] for FakeBalancedGate. When > 0, uses
@@ -210,7 +210,7 @@ class BackendConfig:
     dispatcher_num_sms: int = 20
     dispatcher_share_token_dispatcher: bool = True
     dispatcher_async_dispatch: bool = False
-    enable_deepep: bool | None = None  # Deprecated: use dispatcher="deepep" instead
+    enable_deepep: bool | None = None  # Removed: ignored with a warning; set dispatcher/experts explicitly
     fake_balanced_gate: bool = False
     # Approximate max/mean load ratios (64 experts, top-8, 4096 tokens):
     # 0.0→1.00x, 0.1→~1.2x, 0.3→~1.6x, 0.5→~2.0x, 1.0→~2.8x.
@@ -235,21 +235,17 @@ def __post_init__(self):
         if isinstance(self.gate_precision, str):
             self.gate_precision = dtype_from_str(self.gate_precision, default=None)
 
-        # Handle deprecated enable_deepep parameter
+        # enable_deepep was removed. It is no longer honored; warn (once, on rank 0) if a stale
+        # config still sets it so the user migrates to explicit dispatcher/experts. The field is
+        # retained only so loading an old config does not crash this kw_only dataclass.
         if self.enable_deepep is not None:
-            warnings.warn(
-                "enable_deepep is deprecated and will be removed in a future release. "
-                "Use experts='gmm' and dispatcher='deepep' instead of enable_deepep=True, "
-                "or dispatcher='torch' instead of enable_deepep=False.",
-                DeprecationWarning,
-                stacklevel=2,
-            )
-            if self.enable_deepep:
-                self.experts = "gmm"
-                self.dispatcher = "deepep"
-            else:
-                self.dispatcher = "torch"
-            # Clear the deprecated field after conversion
+            if not torch.distributed.is_initialized() or torch.distributed.get_rank() == 0:
+                logger.warning(
+                    "enable_deepep is no longer supported and is ignored. "
+                    "Set 'dispatcher' (deepep/hybridep/torch) and 'experts' explicitly instead. "
+                    "Previously enable_deepep=True was equivalent to experts=gmm + dispatcher=deepep, "
+                    "and enable_deepep=False to dispatcher=torch."
+                )
             self.enable_deepep = None
 
         # Backward compatibility
diff --git a/tests/functional_tests/hf_transformer_finetune/nemotron_nano_v3_4layer_custom.yaml b/tests/functional_tests/hf_transformer_finetune/nemotron_nano_v3_4layer_custom.yaml
index 884e33f264..d4e62ce36b 100644
--- a/tests/functional_tests/hf_transformer_finetune/nemotron_nano_v3_4layer_custom.yaml
+++ b/tests/functional_tests/hf_transformer_finetune/nemotron_nano_v3_4layer_custom.yaml
@@ -36,7 +36,7 @@ model:
     trust_remote_code: true
   backend:
     _target_: nemo_automodel.components.models.common.BackendConfig
-    enable_deepep: false
+    dispatcher: torch
   
 checkpoint:
   enabled: false
diff --git a/tests/unit_tests/models/deepseek_v32/test_dsv32_state_dict_adapter.py b/tests/unit_tests/models/deepseek_v32/test_dsv32_state_dict_adapter.py
index 6cfafcab61..b19f404883 100644
--- a/tests/unit_tests/models/deepseek_v32/test_dsv32_state_dict_adapter.py
+++ b/tests/unit_tests/models/deepseek_v32/test_dsv32_state_dict_adapter.py
@@ -12,27 +12,27 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import importlib.util
 import sys
 import types
-import importlib.util
-import pytest
+from unittest.mock import Mock, patch
+
 import torch
-from unittest.mock import Mock, patch, MagicMock
 
 # Mock fast_hadamard_transform before importing deepseek_v32 modules
 try:
     import fast_hadamard_transform  # noqa: F401
 except ImportError:
-    if 'fast_hadamard_transform' not in sys.modules:
-        mock_hadamard = types.ModuleType('fast_hadamard_transform')
-        mock_hadamard.__spec__ = importlib.util.spec_from_loader('fast_hadamard_transform', loader=None)
+    if "fast_hadamard_transform" not in sys.modules:
+        mock_hadamard = types.ModuleType("fast_hadamard_transform")
+        mock_hadamard.__spec__ = importlib.util.spec_from_loader("fast_hadamard_transform", loader=None)
         mock_hadamard.hadamard_transform = lambda x, scale: x
-        sys.modules['fast_hadamard_transform'] = mock_hadamard
+        sys.modules["fast_hadamard_transform"] = mock_hadamard
 
+from nemo_automodel.components.models.common import BackendConfig
 from nemo_automodel.components.models.deepseek_v32.config import DeepseekV32Config
 from nemo_automodel.components.models.deepseek_v32.state_dict_adapter import DeepSeekV32StateDictAdapter
 from nemo_automodel.components.moe.config import MoEConfig
-from nemo_automodel.components.models.common import BackendConfig
 
 
 class TestDeepSeekV32StateDictAdapter:
@@ -62,7 +62,7 @@ def create_mock_moe_config(self, **overrides):
 
     def create_mock_backend_config(self, **overrides):
         backend = Mock(spec=BackendConfig)
-        backend.enable_deepep = False
+        backend.dispatcher = "torch"
 
         for key, value in overrides.items():
             setattr(backend, key, value)
@@ -76,10 +76,7 @@ def test_initialization(self):
         backend = self.create_mock_backend_config()
 
         adapter = DeepSeekV32StateDictAdapter(
-            config=config,
-            moe_config=moe_config,
-            backend=backend,
-            dtype=torch.float16
+            config=config, moe_config=moe_config, backend=backend, dtype=torch.float16
         )
 
         assert adapter.config == config
@@ -138,7 +135,7 @@ def create_mock_moe_config(self, **overrides):
 
     def create_mock_backend_config(self, **overrides):
         backend = Mock(spec=BackendConfig)
-        backend.enable_deepep = False
+        backend.dispatcher = "torch"
 
         for key, value in overrides.items():
             setattr(backend, key, value)
@@ -156,7 +153,7 @@ def test_convert_tensor_quantization_normal_weight(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.q_a_proj.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn, tensor, quantization=True)
 
             assert len(result) == 2
@@ -176,7 +173,7 @@ def test_convert_tensor_quantization_skips_layernorm(self):
         tensor = torch.randn(256)
         fqn = "model.layers.0.input_layernorm.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn, tensor, quantization=True)
 
             assert len(result) == 1
@@ -195,7 +192,7 @@ def test_convert_tensor_quantization_skips_indexer_k_norm(self):
         tensor_weight = torch.randn(64)
         fqn_weight = "model.layers.0.self_attn.indexer.k_norm.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn_weight, tensor_weight, quantization=True)
 
             assert len(result) == 1
@@ -206,7 +203,7 @@ def test_convert_tensor_quantization_skips_indexer_k_norm(self):
         tensor_bias = torch.randn(64)
         fqn_bias = "model.layers.0.self_attn.indexer.k_norm.bias"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn_bias, tensor_bias, quantization=True)
 
             assert len(result) == 1
@@ -225,7 +222,7 @@ def test_convert_tensor_quantization_indexer_linear_weights(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.indexer.wq_b.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn, tensor, quantization=True)
 
             assert len(result) == 2
@@ -237,7 +234,7 @@ def test_convert_tensor_quantization_indexer_linear_weights(self):
         tensor_wk = torch.randn(256, 128)
         fqn_wk = "model.layers.0.self_attn.indexer.wk.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn_wk, tensor_wk, quantization=True)
 
             assert len(result) == 2
@@ -257,7 +254,7 @@ def test_convert_tensor_quantization_skips_indexer_weights_proj(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.indexer.weights_proj.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn, tensor, quantization=True)
 
             assert len(result) == 1
@@ -292,7 +289,7 @@ def create_mock_moe_config(self, **overrides):
 
     def create_mock_backend_config(self, **overrides):
         backend = Mock(spec=BackendConfig)
-        backend.enable_deepep = False
+        backend.dispatcher = "torch"
 
         for key, value in overrides.items():
             setattr(backend, key, value)
@@ -394,7 +391,7 @@ def create_mock_moe_config(self, **overrides):
 
     def create_mock_backend_config(self, **overrides):
         backend = Mock(spec=BackendConfig)
-        backend.enable_deepep = False
+        backend.dispatcher = "torch"
 
         for key, value in overrides.items():
             setattr(backend, key, value)
@@ -412,7 +409,7 @@ def test_convert_tensor_with_exclude_regex(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.q_a_proj.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             # With regex that matches the key
             result = adapter.convert_single_tensor_to_hf(
                 fqn, tensor, quantization=False, exclude_key_regex=r".*q_a_proj.*"
@@ -432,7 +429,7 @@ def test_convert_tensor_with_exclude_regex_no_match(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.q_a_proj.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             # With regex that doesn't match the key
             result = adapter.convert_single_tensor_to_hf(
                 fqn, tensor, quantization=False, exclude_key_regex=r".*kv_proj.*"
@@ -453,7 +450,7 @@ def test_convert_tensor_without_quantization(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.q_a_proj.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn, tensor, quantization=False)
 
             # Without quantization, should just return the tensor
@@ -491,7 +488,7 @@ def create_mock_moe_config(self, **overrides):
 
     def create_mock_backend_config(self, **overrides):
         backend = Mock(spec=BackendConfig)
-        backend.enable_deepep = False
+        backend.dispatcher = "torch"
 
         for key, value in overrides.items():
             setattr(backend, key, value)
@@ -555,7 +552,7 @@ def create_mock_moe_config(self, **overrides):
 
     def create_mock_backend_config(self, **overrides):
         backend = Mock(spec=BackendConfig)
-        backend.enable_deepep = False
+        backend.dispatcher = "torch"
 
         for key, value in overrides.items():
             setattr(backend, key, value)
@@ -565,8 +562,8 @@ def create_mock_backend_config(self, **overrides):
     def test_scale_shape_calculation(self):
         """Test that scale shape is calculated correctly."""
         from nemo_automodel.components.models.deepseek_v3.state_dict_adapter import (
-            calculate_scale_shape,
             BLOCK_SIZE,
+            calculate_scale_shape,
         )
 
         # Test with tensor that divides evenly by block size
@@ -588,7 +585,7 @@ def test_quantized_weights_have_correct_scale_shape(self):
         tensor = torch.randn(256, 128)
         fqn = "model.layers.0.self_attn.q_a_proj.weight"
 
-        with patch.object(adapter, '_convert_single_merged_expert_to_hf_split_experts', return_value=None):
+        with patch.object(adapter, "_convert_single_merged_expert_to_hf_split_experts", return_value=None):
             result = adapter.convert_single_tensor_to_hf(fqn, tensor, quantization=True)
 
             # Should have weight and scale_inv
diff --git a/tests/unit_tests/models/minimax_m2/test_minimax_m2_layers.py b/tests/unit_tests/models/minimax_m2/test_minimax_m2_layers.py
index 90212fe05e..24fc3bb18a 100644
--- a/tests/unit_tests/models/minimax_m2/test_minimax_m2_layers.py
+++ b/tests/unit_tests/models/minimax_m2/test_minimax_m2_layers.py
@@ -31,7 +31,7 @@ def backend():
         attn="sdpa",
         rms_norm="torch",
         rope_fusion=False,
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/minimax_m2/test_minimax_m2_model.py b/tests/unit_tests/models/minimax_m2/test_minimax_m2_model.py
index 35f1b3785f..ccfe19663c 100644
--- a/tests/unit_tests/models/minimax_m2/test_minimax_m2_model.py
+++ b/tests/unit_tests/models/minimax_m2/test_minimax_m2_model.py
@@ -49,7 +49,7 @@ def backend():
         attn="sdpa",
         rms_norm="torch",
         rope_fusion=False,
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter.py b/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter.py
index 13597b0bfc..80f7f014b8 100644
--- a/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter.py
+++ b/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter.py
@@ -54,7 +54,7 @@ def backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter_dtensor.py b/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter_dtensor.py
index 7c12685891..f3cdd0bd89 100644
--- a/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter_dtensor.py
+++ b/tests/unit_tests/models/minimax_m2/test_minimax_m2_state_dict_adapter_dtensor.py
@@ -46,7 +46,7 @@ def adapter():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_layers.py b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_layers.py
index e73aabc5b8..8d316139df 100644
--- a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_layers.py
+++ b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_layers.py
@@ -216,7 +216,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=False,
             enable_hf_state_dict_adapter=False,
         )
@@ -517,7 +517,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=False,
             enable_hf_state_dict_adapter=False,
         )
diff --git a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_model.py b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_model.py
index ed31d1e459..85c45aa31d 100644
--- a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_model.py
+++ b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_model.py
@@ -97,7 +97,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=False,
             enable_hf_state_dict_adapter=False,
         )
@@ -328,7 +328,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=False,
             enable_hf_state_dict_adapter=False,
         )
@@ -762,7 +762,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=False,
             enable_hf_state_dict_adapter=False,
         )
@@ -960,7 +960,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=False,
             enable_hf_state_dict_adapter=False,
         )
@@ -1118,7 +1118,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
             fake_balanced_gate=True,  # Use fake balanced gate for deterministic testing
             enable_hf_state_dict_adapter=False,
         )
diff --git a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp.py b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp.py
index 4c4ffecdf5..2f6644e02d 100644
--- a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp.py
+++ b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp.py
@@ -223,7 +223,7 @@ def backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=True,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp_parity.py b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp_parity.py
index 14c0151d26..13c1827948 100644
--- a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp_parity.py
+++ b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_mtp_parity.py
@@ -48,7 +48,7 @@ def backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=True,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_pp_mtp.py b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_pp_mtp.py
index 0fdb3fc61d..316cd413a5 100644
--- a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_pp_mtp.py
+++ b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_pp_mtp.py
@@ -37,7 +37,7 @@ def backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=True,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_state_dict_adapter.py b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_state_dict_adapter.py
index 3e7d34e10d..b38d28caf1 100644
--- a/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_state_dict_adapter.py
+++ b/tests/unit_tests/models/nemotron_v3/test_nemotron_v3_state_dict_adapter.py
@@ -74,7 +74,7 @@ def backend(self):
             linear="torch",
             attn="sdpa",
             rms_norm="torch",
-            enable_deepep=False,
+            dispatcher="torch",
         )
 
     def test_adapter_init(self, config, moe_config, backend):
@@ -228,7 +228,7 @@ def test_from_hf_renames_backbone_to_model(self, config, moe_config, backend):
                 "lm_head.weight": torch.randn(100, 256),
             }
 
-            result = adapter.from_hf(hf_state_dict)
+            adapter.from_hf(hf_state_dict)
 
             # Check that _from_hf_w_merged_experts was called with renamed state dict
             call_args = mock_merge.call_args[0][0]
@@ -486,7 +486,10 @@ def test_from_hf_uses_mixer_experts_path(self, config, moe_config, backend):
 
         with patch.object(adapter, "_validate_expert_availability"):
             with patch("nemo_automodel.components.moe.state_dict_mixin.should_load_expert_for_rank", return_value=True):
-                with patch("nemo_automodel.components.moe.state_dict_mixin.create_dtensor_from_local", side_effect=lambda x, *args: x):
+                with patch(
+                    "nemo_automodel.components.moe.state_dict_mixin.create_dtensor_from_local",
+                    side_effect=lambda x, *args: x,
+                ):
                     result = adapter._from_hf_w_merged_experts(hf_state_dict)
 
         # Should have created merged expert tensors with mixer.experts path
diff --git a/tests/unit_tests/models/qwen3_5/test_qwen3_5_mtp.py b/tests/unit_tests/models/qwen3_5/test_qwen3_5_mtp.py
index 9baaf4c377..a7a91d7ab0 100644
--- a/tests/unit_tests/models/qwen3_5/test_qwen3_5_mtp.py
+++ b/tests/unit_tests/models/qwen3_5/test_qwen3_5_mtp.py
@@ -74,7 +74,7 @@ def _backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/models/qwen3_5/test_qwen3_5_pp.py b/tests/unit_tests/models/qwen3_5/test_qwen3_5_pp.py
index 7203041735..5c1c79315c 100644
--- a/tests/unit_tests/models/qwen3_5/test_qwen3_5_pp.py
+++ b/tests/unit_tests/models/qwen3_5/test_qwen3_5_pp.py
@@ -194,7 +194,7 @@ def _tiny_vlm_model():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_model.py b/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_model.py
index d95ad4bc3c..cda01163ea 100644
--- a/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_model.py
+++ b/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_model.py
@@ -129,7 +129,7 @@ def backend_config():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_mtp.py b/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_mtp.py
index 2d03e8ed8e..71287b5fdb 100644
--- a/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_mtp.py
+++ b/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_mtp.py
@@ -66,7 +66,7 @@ def _backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_state_dict_adapter.py b/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_state_dict_adapter.py
index 77fa1cd85b..e936ce3810 100644
--- a/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_state_dict_adapter.py
+++ b/tests/unit_tests/models/qwen3_5_moe/test_qwen3_5_moe_state_dict_adapter.py
@@ -70,7 +70,7 @@ def backend_config():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/step3p5/test_step3p5_layers.py b/tests/unit_tests/models/step3p5/test_step3p5_layers.py
index 8b4f56f076..906d8dc8d0 100644
--- a/tests/unit_tests/models/step3p5/test_step3p5_layers.py
+++ b/tests/unit_tests/models/step3p5/test_step3p5_layers.py
@@ -26,13 +26,13 @@
     Step3p5RotaryEmbedding,
 )
 
-
 pytestmark = pytest.mark.skipif(not torch.cuda.is_available(), reason="CUDA not available")
 
 
 @dataclass
 class MockStep3p5Config:
     """Mock configuration for Step3p5 model."""
+
     vocab_size: int = 128
     hidden_size: int = 64
     intermediate_size: int = 128
@@ -74,7 +74,7 @@ def sdpa_backend():
         attn="sdpa",
         rms_norm="torch",
         rope_fusion=False,
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=False,
     )
@@ -277,7 +277,10 @@ def test_forward_shape_preserved(self, config, sdpa_backend):
         fake_attn = torch.zeros(batch, config.num_attention_heads, seq, config.head_dim)
         attention.attn_func = MagicMock(return_value=fake_attn.to(torch.bfloat16))
 
-        with patch("nemo_automodel.components.models.step3p5.layers.apply_rotary_emb_qk", side_effect=lambda q, k, *_, **__: (q, k)):
+        with patch(
+            "nemo_automodel.components.models.step3p5.layers.apply_rotary_emb_qk",
+            side_effect=lambda q, k, *_, **__: (q, k),
+        ):
             out = attention(x, freqs_cis=freqs_cis)
 
         assert out.shape == (batch, seq, config.hidden_size)
diff --git a/tests/unit_tests/models/step3p5/test_step3p5_model.py b/tests/unit_tests/models/step3p5/test_step3p5_model.py
index 0d5e61f1c5..fcf6a2b278 100644
--- a/tests/unit_tests/models/step3p5/test_step3p5_model.py
+++ b/tests/unit_tests/models/step3p5/test_step3p5_model.py
@@ -92,7 +92,7 @@ def sdpa_backend():
         attn="sdpa",
         rms_norm="torch",
         rope_fusion=False,
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=False,
     )
diff --git a/tests/unit_tests/models/step3p5/test_step3p5_state_dict_adapter.py b/tests/unit_tests/models/step3p5/test_step3p5_state_dict_adapter.py
index fb0c814518..0a5b699b87 100644
--- a/tests/unit_tests/models/step3p5/test_step3p5_state_dict_adapter.py
+++ b/tests/unit_tests/models/step3p5/test_step3p5_state_dict_adapter.py
@@ -70,7 +70,7 @@ def backend():
         linear="torch",
         attn="sdpa",
         rms_norm="torch",
-        enable_deepep=False,
+        dispatcher="torch",
         fake_balanced_gate=False,
         enable_hf_state_dict_adapter=True,
     )
diff --git a/tests/unit_tests/moe/test_backend_config.py b/tests/unit_tests/moe/test_backend_config.py
index 529370cb40..a617239020 100644
--- a/tests/unit_tests/moe/test_backend_config.py
+++ b/tests/unit_tests/moe/test_backend_config.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import warnings
+import logging
 
 import pytest
 import torch
@@ -122,66 +122,43 @@ def test_fake_gate_noise_with_fake_balanced_gate(self):
         assert config.fake_gate_noise == 0.3
 
 
-class TestBackendConfigEnableDeepepDeprecation:
-    """Test backwards compatibility for deprecated enable_deepep parameter."""
-
-    def test_enable_deepep_true_sets_dispatcher_deepep_and_experts_gmm(self):
-        """Test that enable_deepep=True sets dispatcher='deepep' and experts='gmm' with deprecation warning."""
-        with warnings.catch_warnings(record=True) as w:
-            warnings.simplefilter("always")
-            config = BackendConfig(enable_deepep=True)
-            assert config.dispatcher == "deepep"
-            assert config.experts == "gmm"
-            assert config.enable_deepep is None  # Should be cleared after conversion
-            assert len(w) == 1
-            assert issubclass(w[0].category, DeprecationWarning)
-            assert "enable_deepep is deprecated" in str(w[0].message)
-
-    def test_enable_deepep_false_sets_dispatcher_torch(self):
-        """Test that enable_deepep=False sets dispatcher='torch' with deprecation warning."""
-        with warnings.catch_warnings(record=True) as w:
-            warnings.simplefilter("always")
-            config = BackendConfig(enable_deepep=False)
-            assert config.dispatcher == "torch"
-            expected_experts = "torch_mm" if torch.cuda.is_available() else "torch"
-            assert config.experts == expected_experts  # experts unchanged when enable_deepep=False
-            assert config.enable_deepep is None  # Should be cleared after conversion
-            assert len(w) == 1
-            assert issubclass(w[0].category, DeprecationWarning)
-            assert "enable_deepep is deprecated" in str(w[0].message)
-
-    def test_enable_deepep_none_no_warning(self):
-        """Test that enable_deepep=None (default) does not trigger warning."""
-        with warnings.catch_warnings(record=True) as w:
-            warnings.simplefilter("always")
+class TestBackendConfigEnableDeepepRemoved:
+    """enable_deepep was removed: it is ignored (with a warning) and never alters dispatcher/experts."""
+
+    def test_enable_deepep_true_is_ignored_and_warns(self, caplog):
+        """enable_deepep=True is ignored; dispatcher/experts keep their explicit values and a warning is logged."""
+        with caplog.at_level(logging.WARNING):
+            config = BackendConfig(dispatcher="hybridep", experts="gmm", enable_deepep=True)
+        assert config.dispatcher == "hybridep"  # not overridden to "deepep"
+        assert config.experts == "gmm"
+        assert config.enable_deepep is None  # cleared after the warning
+        assert "enable_deepep is no longer supported" in caplog.text
+
+    def test_enable_deepep_false_is_ignored_and_warns(self, caplog):
+        """enable_deepep=False is ignored; the dispatcher is NOT forced to torch and a warning is logged."""
+        with caplog.at_level(logging.WARNING):
+            config = BackendConfig(dispatcher="deepep", experts="gmm", enable_deepep=False)
+        assert config.dispatcher == "deepep"  # not forced to "torch"
+        assert config.experts == "gmm"
+        assert config.enable_deepep is None
+        assert "enable_deepep is no longer supported" in caplog.text
+
+    def test_enable_deepep_none_no_warning(self, caplog):
+        """enable_deepep=None (default) leaves the field as None and logs no warning."""
+        with caplog.at_level(logging.WARNING):
             config = BackendConfig()
-            assert config.enable_deepep is None
-            # No deprecation warning should be raised
-            deprecation_warnings = [x for x in w if issubclass(x.category, DeprecationWarning)]
-            assert len(deprecation_warnings) == 0
-
-    def test_enable_deepep_overrides_dispatcher_and_experts(self):
-        """Test that enable_deepep takes precedence over dispatcher and experts when both provided."""
-        with warnings.catch_warnings(record=True) as w:
-            warnings.simplefilter("always")
-            # Even if dispatcher="torch" and experts="torch", enable_deepep=True should override them
+        assert config.enable_deepep is None
+        assert "enable_deepep" not in caplog.text
+
+    def test_enable_deepep_does_not_override_explicit_dispatcher(self, caplog):
+        """A stale enable_deepep no longer wins over an explicit dispatcher/experts."""
+        with caplog.at_level(logging.WARNING):
             config = BackendConfig(dispatcher="torch", experts="torch", enable_deepep=True)
-            assert config.dispatcher == "deepep"
-            assert config.experts == "gmm"
-            assert len(w) == 1
-            assert issubclass(w[0].category, DeprecationWarning)
-
-    def test_enable_deepep_false_overrides_dispatcher_deepep(self):
-        """Test that enable_deepep=False overrides dispatcher='deepep'."""
-        with warnings.catch_warnings(record=True) as w:
-            warnings.simplefilter("always")
-            config = BackendConfig(dispatcher="deepep", enable_deepep=False)
-            assert config.dispatcher == "torch"
-            assert len(w) == 1
-            assert issubclass(w[0].category, DeprecationWarning)
+        assert config.dispatcher == "torch"
+        assert config.experts == "torch"
 
     def test_dispatcher_without_enable_deepep(self):
-        """Test that dispatcher works correctly without enable_deepep."""
+        """dispatcher works correctly without enable_deepep (field stays None)."""
         config = BackendConfig(dispatcher="deepep")
         assert config.dispatcher == "deepep"
         assert config.enable_deepep is None
@@ -190,17 +167,6 @@ def test_dispatcher_without_enable_deepep(self):
         assert config.dispatcher == "torch"
         assert config.enable_deepep is None
 
-    def test_deprecation_warning_message_content(self):
-        """Test that deprecation warning message contains helpful migration info."""
-        with warnings.catch_warnings(record=True) as w:
-            warnings.simplefilter("always")
-            BackendConfig(enable_deepep=True)
-            warning_message = str(w[0].message)
-            assert "experts='gmm'" in warning_message
-            assert "dispatcher='deepep'" in warning_message
-            assert "dispatcher='torch'" in warning_message
-            assert "will be removed in a future release" in warning_message
-
 
 class TestBackendConfigHybridEP:
     """Test BackendConfig HybridEP dispatcher support."""