Address review

Signed-off-by: Hollow Man <hollowman@opensuse.org>

Author: HollowMan6

src/megatron/bridge/perf_recipes/glm_moe_dsa/common.py

@@ -21,8 +21,6 @@
 from megatron.bridge.training.config import ConfigContainer
 
 
-GLM51_MODEL_ID = "zai-org/GLM-5.1"
-GLM52_MODEL_ID = "zai-org/GLM-5.2"
 GLM5_LONG_CONTEXT = 131072
 
 

src/megatron/bridge/perf_recipes/glm_moe_dsa/gb200/glm5.py

@@ -14,8 +14,6 @@
 """GB200 performance recipes for GLM-5.1 and GLM-5.2 SFT."""
 
 from megatron.bridge.perf_recipes.glm_moe_dsa.common import (
-    GLM51_MODEL_ID,
-    GLM52_MODEL_ID,
     ConfigContainer,
     _glm5_cudnn_sft_base,
 )
@@ -41,9 +39,9 @@ def _glm5_gb200_cudnn_sft_config(model_id: str) -> ConfigContainer:
 
 def glm51_sft_192gpu_gb200_bf16_config() -> ConfigContainer:
     """GLM-5.1 SFT: 192× GB200, BF16, 128K packed THD, CP=32, cuDNN DSA."""
-    return _glm5_gb200_cudnn_sft_config(GLM51_MODEL_ID)
+    return _glm5_gb200_cudnn_sft_config("zai-org/GLM-5.1")
 
 
 def glm52_sft_192gpu_gb200_bf16_config() -> ConfigContainer:
     """GLM-5.2 SFT: 192× GB200, BF16, 128K packed THD, CP=32, cuDNN DSA."""
-    return _glm5_gb200_cudnn_sft_config(GLM52_MODEL_ID)
+    return _glm5_gb200_cudnn_sft_config("zai-org/GLM-5.2")

src/megatron/bridge/perf_recipes/glm_moe_dsa/h100/glm5.py

@@ -14,8 +14,6 @@
 """H100 performance recipes for GLM-5.1 and GLM-5.2 SFT."""
 
 from megatron.bridge.perf_recipes.glm_moe_dsa.common import (
-    GLM51_MODEL_ID,
-    GLM52_MODEL_ID,
     ConfigContainer,
     _glm5_cudnn_sft_base,
 )
@@ -28,10 +26,10 @@
 _GLM5_H100_GBS = 520
 
 
-def _glm5_h100_cudnn_sft_config(model_id: str) -> ConfigContainer:
-    """Return the 52-node H100 GLM5 cuDNN SFT benchmark shape."""
+def glm51_sft_416gpu_h100_bf16_config() -> ConfigContainer:
+    """GLM-5.1 SFT: 416x H100, BF16, 128K packed THD, CP=4, cuDNN DSA."""
     return _glm5_cudnn_sft_base(
-        model_id,
+        "zai-org/GLM-5.1",
         tensor_model_parallel_size=_GLM5_H100_TP,
         pipeline_model_parallel_size=_GLM5_H100_PP,
         context_parallel_size=_GLM5_H100_CP,
@@ -41,11 +39,14 @@ def _glm5_h100_cudnn_sft_config(model_id: str) -> ConfigContainer:
     )
 
 
-def glm51_sft_416gpu_h100_bf16_config() -> ConfigContainer:
-    """GLM-5.1 SFT: 416x H100, BF16, 128K packed THD, CP=4, cuDNN DSA."""
-    return _glm5_h100_cudnn_sft_config(GLM51_MODEL_ID)
-
-
 def glm52_sft_416gpu_h100_bf16_config() -> ConfigContainer:
     """GLM-5.2 SFT: 416x H100, BF16, 128K packed THD, CP=4, cuDNN DSA."""
-    return _glm5_h100_cudnn_sft_config(GLM52_MODEL_ID)
+    return _glm5_cudnn_sft_base(
+        "zai-org/GLM-5.2",
+        tensor_model_parallel_size=_GLM5_H100_TP,
+        pipeline_model_parallel_size=_GLM5_H100_PP,
+        context_parallel_size=_GLM5_H100_CP,
+        expert_model_parallel_size=_GLM5_H100_EP,
+        global_batch_size=_GLM5_H100_GBS,
+        sequence_parallel=True,
+    )