fix: add rotary_percent to Megatron TransformerConfig patch so MiniMax-M2 hf->torch_dist conversion works

docker/patch/latest/megatron.patch

@@ -862,10 +862,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index dce438520..de51edaf3 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -229,6 +229,9 @@ class TransformerConfig(ModelParallelConfig):
+@@ -229,6 +229,12 @@ class TransformerConfig(ModelParallelConfig):
      attention_output_gate: bool = False
      """Whether to apply output gate to the attention layers."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +

docker/patch/v0.5.0rc0-cu126/megatron.patch

@@ -329,10 +329,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index 6f557e1f..b295fd35 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -173,6 +173,9 @@ class TransformerConfig(ModelParallelConfig):
+@@ -173,6 +173,12 @@ class TransformerConfig(ModelParallelConfig):
      qk_layernorm: bool = False
      """Whether to apply `normalization` type of normalization to the query and key embeddings."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +

docker/patch/v0.5.12.post1/megatron.patch

@@ -862,10 +862,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index dce438520..de51edaf3 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -229,6 +229,9 @@ class TransformerConfig(ModelParallelConfig):
+@@ -229,6 +229,12 @@ class TransformerConfig(ModelParallelConfig):
      attention_output_gate: bool = False
      """Whether to apply output gate to the attention layers."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +

docker/patch/v0.5.5.post1/megatron.patch

@@ -723,10 +723,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index d55bebe7e..1eecbbd38 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -173,6 +173,10 @@ class TransformerConfig(ModelParallelConfig):
+@@ -173,6 +173,13 @@ class TransformerConfig(ModelParallelConfig):
      qk_layernorm: bool = False
      """Whether to apply `normalization` type of normalization to the query and key embeddings."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +    use_gated_attention: bool = False

docker/patch/v0.5.6/megatron.patch

@@ -736,10 +736,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index d55bebe7e..1eecbbd38 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -173,6 +173,10 @@ class TransformerConfig(ModelParallelConfig):
+@@ -173,6 +173,13 @@ class TransformerConfig(ModelParallelConfig):
      qk_layernorm: bool = False
      """Whether to apply `normalization` type of normalization to the query and key embeddings."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +    use_gated_attention: bool = False

docker/patch/v0.5.7/megatron.patch

@@ -650,10 +650,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index e2705bd9f..a0aa109b5 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -210,6 +210,9 @@ class TransformerConfig(ModelParallelConfig):
+@@ -210,6 +210,12 @@ class TransformerConfig(ModelParallelConfig):
      attention_output_gate: bool = False
      """Whether to apply output gate to the attention layers."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +

docker/patch/v0.5.9/megatron.patch

@@ -673,10 +673,13 @@ diff --git a/megatron/core/transformer/transformer_config.py b/megatron/core/tra
 index e2705bd9f..a0aa109b5 100644
 --- a/megatron/core/transformer/transformer_config.py
 +++ b/megatron/core/transformer/transformer_config.py
-@@ -210,6 +210,9 @@ class TransformerConfig(ModelParallelConfig):
+@@ -210,6 +210,12 @@ class TransformerConfig(ModelParallelConfig):
      attention_output_gate: bool = False
      """Whether to apply output gate to the attention layers."""
 
++    rotary_percent: Optional[float] = None
++    """Percent of rotary dimension to use for rotary position embeddings."""
++
 +    post_self_attn_layernorm: bool = False
 +    post_mlp_layernorm: bool = False
 +