fix(recipe): give Moonlight-16B 8-chunk pipeline layouts 27 decoder layers (#6359)

Signed-off-by: Pruthviraj Prakash <pruprakash@nvidia.com>

Author: pruprakash

src/megatron/bridge/recipes/moonlight/h100/moonlight_16b.py

@@ -138,9 +138,15 @@ def _get_moonlight_pipeline_layout(pp_size: int, vp_size: int | None):
         (1, 1): None,
         (2, 1): [["embedding"] + ["decoder"] * 14, ["decoder"] * 13 + ["loss"]],
         (4, 1): [["embedding"] + ["decoder"] * 7] + [["decoder"] * 7] * 2 + [["decoder"] * 6 + ["loss"]],
-        (8, 1): [["embedding"] + ["decoder"] * 4] + [["decoder"] * 4] * 6 + [["decoder"] * 3 + ["loss"]],
+        (8, 1): [["embedding"] + ["decoder"] * 4]
+        + [["decoder"] * 4] * 2
+        + [["decoder"] * 3] * 4
+        + [["decoder"] * 3 + ["loss"]],
         (2, 2): [["embedding"] + ["decoder"] * 7] + [["decoder"] * 7] * 2 + [["decoder"] * 6 + ["loss"]],
-        (4, 2): [["embedding"] + ["decoder"] * 4] + [["decoder"] * 4] * 6 + [["decoder"] * 3 + ["loss"]],
+        (4, 2): [["embedding"] + ["decoder"] * 4]
+        + [["decoder"] * 4] * 2
+        + [["decoder"] * 3] * 4
+        + [["decoder"] * 3 + ["loss"]],
     }
     vp_size = 1 if vp_size is None else vp_size
     if (pp_size, vp_size) not in map_pp_vp_to_layout:

tests/unit_tests/recipes/test_moonlight_recipes.py

@@ -531,6 +531,38 @@ def test_moonlight_pipeline_layout_tracks_supported_runner_override(monkeypatch:
     assert cfg.model.moe_hybridep_pad_uneven_dispatch_inputs is True
 
 
+_MOONLIGHT_16B_NUM_LAYERS = 27
+
+
+def _supported_moonlight_pp_vp_layouts():
+    """Return every (pp, vp, layout) the Moonlight-16B layout helper supports."""
+    from megatron.bridge.recipes.moonlight.h100.moonlight_16b import _get_moonlight_pipeline_layout
+
+    supported = []
+    for pp in range(1, 17):
+        for vp in range(1, 9):
+            try:
+                layout = _get_moonlight_pipeline_layout(pp, vp)
+            except ValueError:
+                continue
+            if layout is not None:
+                supported.append((pp, vp, layout))
+    return supported
+
+
+def test_moonlight_pipeline_layouts_match_model_depth():
+    """Every supported Moonlight-16B pipeline layout must pass MCore layout validation."""
+    from megatron.core.transformer.pipeline_parallel_layer_layout import PipelineParallelLayerLayout
+
+    supported = _supported_moonlight_pp_vp_layouts()
+    assert supported
+
+    for pp, vp, layout in supported:
+        parsed = PipelineParallelLayerLayout(layout, pipeline_model_parallel_size=pp)
+        assert parsed.virtual_pipeline_model_parallel_size == vp, (pp, vp)
+        parsed.validate_layer_layout(num_layers=_MOONLIGHT_16B_NUM_LAYERS, mtp_num_layers=None)
+
+
 def test_moonlight_16b_peft_convergence_contract(monkeypatch: pytest.MonkeyPatch):
     """Test the exact 4-GPU LoRA convergence contract and frozen base."""
     from megatron.bridge.peft.lora import LoRA