[model, refactor] refactor: Centralize provider_bridge config mapping in base class

src/megatron/bridge/models/deepseek/deepseek_provider.py

@@ -13,110 +13,57 @@
 # limitations under the License.
 import warnings
 from dataclasses import dataclass, field
-from functools import partial
-from typing import TYPE_CHECKING, Callable, List, Optional, Union
+from typing import Callable, List, Optional, Union
 
 import torch
 import torch.nn.functional as F
-from megatron.core.models.gpt.gpt_layer_specs import get_gpt_decoder_block_spec
 
-from megatron.bridge.models.gpt_provider import GPTModelProvider
-from megatron.bridge.models.transformer_config import MLATransformerConfig
+from megatron.bridge.models.mla_provider import MLAModelProvider
 from megatron.bridge.utils.common_utils import get_rank_safe
 
 
-try:
-    import transformer_engine  # type: ignore  # noqa: F401
-
-    HAVE_TE = True
-except (ImportError, ModuleNotFoundError):
-    HAVE_TE = False
-
-if TYPE_CHECKING:
-    from megatron.core.transformer import ModuleSpec
-
-if HAVE_TE:
-    from megatron.core.utils import is_te_min_version
+def _warn_deprecated(old_cls: str, new_cls: str = "MLAModelProvider") -> None:
+    if get_rank_safe() == 0:
+        warnings.warn(
+            f"{old_cls} is deprecated and will be removed in a future release. "
+            f"Use {new_cls} with MEGATRON_DEFAULTS in the bridge instead.",
+            DeprecationWarning,
+            stacklevel=3,
+        )
 
 
 @dataclass
-class DeepSeekModelProvider(MLATransformerConfig, GPTModelProvider):
-    """
-    Base config for DeepSeek V2 and V3 models.
-    """
+class DeepSeekModelProvider(MLAModelProvider):
+    """Deprecated alias for ``MLAModelProvider``.
 
-    transformer_layer_spec: Union["ModuleSpec", Callable[["GPTModelProvider"], "ModuleSpec"]] = partial(
-        get_gpt_decoder_block_spec, use_transformer_engine=HAVE_TE
-    )
+    Deprecated:
+        This alias remains for backward compatibility and will be removed in a
+        future release. Use ``MLAModelProvider`` instead.
+    """
 
-    # Model
+    # Common DeepSeek defaults
     normalization: str = "RMSNorm"
     activation_func: Callable = F.silu
-    gated_linear_unit: bool = True  # swiglu
+    gated_linear_unit: bool = True
     position_embedding_type: str = "rope"
     add_bias_linear: bool = False
     share_embeddings_and_output_weights: bool = False
-    num_attention_heads: int = 128
-    kv_channels: int = 128
-    max_position_embeddings: int = 4096
-    seq_length: int = 4096
-    rotary_base: float = 10000.0
-    make_vocab_size_divisible_by: int = 3200
-    mtp_num_layers: Optional[int] = None
-    mtp_loss_scaling_factor: Optional[float] = None
-
-    # Regularization
-    attention_dropout: float = 0.0
-    hidden_dropout: float = 0.0
     qk_layernorm: bool = True
-
-    # MoE
+    bf16: bool = True
+    params_dtype: torch.dtype = torch.bfloat16
     moe_grouped_gemm: bool = True
-    moe_router_pre_softmax: bool = True
     moe_token_dispatcher_type: str = "alltoall"
-    moe_router_load_balancing_type: str = "seq_aux_loss"
-    moe_shared_expert_overlap: bool = True
-    moe_router_dtype: Optional[str] = "fp32"
-
-    # MLA
-    q_lora_rank: int = 1536
+    # MLA defaults
+    q_lora_rank: Optional[int] = 1536
     kv_lora_rank: int = 512
-    qk_head_dim: int = 128
-    qk_pos_emb_head_dim: int = 64
-    v_head_dim: int = 128
-    rotary_scaling_factor: float = 40
-    mscale: float = 1.0
-    mscale_all_dim: float = 1.0
-
-    # Miscellaneous
-    init_method_std: float = 0.006
-    layernorm_epsilon: float = 1e-6
-    bf16: bool = True
-    params_dtype: torch.dtype = torch.bfloat16
-    async_tensor_model_parallel_allreduce: bool = True
-    attention_softmax_in_fp32: bool = False
-    persist_layer_norm: bool = True
-    num_layers_in_first_pipeline_stage: Optional[int] = None
-    num_layers_in_last_pipeline_stage: Optional[int] = None
-    account_for_embedding_in_pipeline_split: bool = False
-    account_for_loss_in_pipeline_split: bool = False
 
-    # MLA specific
-    multi_latent_attention: bool = True
-
-    # fusions
-    apply_rope_fusion: bool = False
-    bias_activation_fusion: bool = True
-    bias_dropout_fusion: bool = True
-    masked_softmax_fusion: bool = True
-    gradient_accumulation_fusion: bool = True
-    cross_entropy_loss_fusion: bool = True
-    cross_entropy_fusion_impl: str = "te"
-    moe_permute_fusion: bool = is_te_min_version("2.1.0") if HAVE_TE else False
+    def __post_init__(self) -> None:
+        _warn_deprecated("DeepSeekModelProvider")
+        super().__post_init__()
 
 
 @dataclass
-class DeepSeekV2ModelProvider(DeepSeekModelProvider):
+class DeepSeekV2ModelProvider(MLAModelProvider):
     """
     DeepSeek-V2 Model: https://github.com/deepseek-ai/DeepSeek-V2
     """
@@ -137,9 +84,13 @@ class DeepSeekV2ModelProvider(DeepSeekModelProvider):
     mscale_all_dim: float = 0.707
     vocab_size: int = 102400
 
+    def __post_init__(self) -> None:
+        _warn_deprecated("DeepSeekV2ModelProvider")
+        super().__post_init__()
+
 
 @dataclass
-class DeepSeekV2LiteModelProvider(DeepSeekV2ModelProvider):
+class DeepSeekV2LiteModelProvider(MLAModelProvider):
     """
     DeepSeek-V2-Lite Model: https://github.com/deepseek-ai/DeepSeek-V2
     HuggingFace: https://huggingface.co/deepseek-ai/DeepSeek-V2-Lite
@@ -150,7 +101,7 @@ class DeepSeekV2LiteModelProvider(DeepSeekV2ModelProvider):
     ffn_hidden_size: int = 10944
     num_attention_heads: int = 16
     kv_channels: int = 16
-    q_lora_rank: int = None
+    q_lora_rank: Optional[int] = None
     num_moe_experts: int = 64
     moe_ffn_hidden_size: int = 1408
     moe_shared_expert_intermediate_size: int = 2816  # 1408 * 2 shared experts
@@ -159,18 +110,25 @@ class DeepSeekV2LiteModelProvider(DeepSeekV2ModelProvider):
     moe_router_num_groups: int = 1
     moe_router_group_topk: int = 1
     moe_router_topk_scaling_factor: float = 1.0
+    mscale: float = 0.707
+    mscale_all_dim: float = 0.707
     vocab_size: int = 102400
 
+    def __post_init__(self) -> None:
+        _warn_deprecated("DeepSeekV2LiteModelProvider")
+        super().__post_init__()
+
 
 @dataclass
-class DeepSeekV3ModelProvider(DeepSeekModelProvider):
+class DeepSeekV3ModelProvider(MLAModelProvider):
     """
     DeepSeek-V3 Model: https://github.com/deepseek-ai/DeepSeek-V3
     """
 
     num_layers: int = 61
     hidden_size: int = 7168
     ffn_hidden_size: int = 18432
+    kv_channels: int = 128
     num_moe_experts: int = 256
     moe_ffn_hidden_size: int = 2048
     moe_shared_expert_intermediate_size: int = 2048  # 2048 * 1 shared expert
@@ -190,9 +148,13 @@ class DeepSeekV3ModelProvider(DeepSeekModelProvider):
     mscale_all_dim: float = 1.0
     vocab_size: int = 129280
 
+    def __post_init__(self) -> None:
+        _warn_deprecated("DeepSeekV3ModelProvider")
+        super().__post_init__()
+
 
 @dataclass
-class MoonlightModelProvider16B(DeepSeekModelProvider):
+class MoonlightModelProvider16B(MLAModelProvider):
     """
     Moonlight-16B-A3B Model: https://github.com/moonshotai/Moonlight-16B-A3B
 
@@ -228,86 +190,14 @@ class MoonlightModelProvider16B(DeepSeekModelProvider):
     rotary_percent: float = 1.0
     vocab_size: int = 163842
 
-
-# -----------------------------------------------------------------------------
-# Deprecated aliases (to be removed in a future release)
-# -----------------------------------------------------------------------------
-
-
-def _warn_deprecated(old_cls: str, new_cls: str) -> None:
-    if get_rank_safe() == 0:
-        warnings.warn(
-            f"{old_cls} is deprecated and will be removed in a future release. Use {new_cls} instead.",
-            DeprecationWarning,
-            stacklevel=2,
-        )
-
-
-@dataclass
-class DeepSeekProvider(DeepSeekModelProvider):
-    """Deprecated alias for ``DeepSeekModelProvider``.
-
-    Deprecated:
-        This alias remains for backward compatibility and will be removed in a
-        future release. Import and use ``DeepSeekModelProvider`` instead.
-    """
-
-    def __post_init__(self) -> None:
-        _warn_deprecated("DeepSeekProvider", "DeepSeekModelProvider")
-        super().__post_init__()
-
-
-@dataclass
-class DeepSeekV2Provider(DeepSeekV2ModelProvider):
-    """Deprecated alias for ``DeepSeekV2ModelProvider``.
-
-    Deprecated:
-        This alias remains for backward compatibility and will be removed in a
-        future release. Import and use ``DeepSeekV2ModelProvider`` instead.
-    """
-
-    def __post_init__(self) -> None:
-        _warn_deprecated("DeepSeekV2Provider", "DeepSeekV2ModelProvider")
-        super().__post_init__()
-
-
-@dataclass
-class DeepSeekV2LiteProvider(DeepSeekV2LiteModelProvider):
-    """Deprecated alias for ``DeepSeekV2LiteModelProvider``.
-
-    Deprecated:
-        This alias remains for backward compatibility and will be removed in a
-        future release. Import and use ``DeepSeekV2LiteModelProvider`` instead.
-    """
-
     def __post_init__(self) -> None:
-        _warn_deprecated("DeepSeekV2LiteProvider", "DeepSeekV2LiteModelProvider")
+        _warn_deprecated("MoonlightModelProvider16B")
         super().__post_init__()
 
 
-@dataclass
-class DeepSeekV3Provider(DeepSeekV3ModelProvider):
-    """Deprecated alias for ``DeepSeekV3ModelProvider``.
-
-    Deprecated:
-        This alias remains for backward compatibility and will be removed in a
-        future release. Import and use ``DeepSeekV3ModelProvider`` instead.
-    """
-
-    def __post_init__(self) -> None:
-        _warn_deprecated("DeepSeekV3Provider", "DeepSeekV3ModelProvider")
-        super().__post_init__()
-
-
-@dataclass
-class MoonlightProvider(MoonlightModelProvider16B):
-    """Deprecated alias for ``MoonlightModelProvider16B``.
-
-    Deprecated:
-        This alias remains for backward compatibility and will be removed in a
-        future release. Import and use ``MoonlightModelProvider16B`` instead.
-    """
-
-    def __post_init__(self) -> None:
-        _warn_deprecated("MoonlightProvider", "MoonlightModelProvider16B")
-        super().__post_init__()
+# Legacy aliases for backward compatibility
+DeepSeekProvider = DeepSeekModelProvider
+DeepSeekV2Provider = DeepSeekV2ModelProvider
+DeepSeekV2LiteProvider = DeepSeekV2LiteModelProvider
+DeepSeekV3Provider = DeepSeekV3ModelProvider
+MoonlightProvider = MoonlightModelProvider16B

src/megatron/bridge/models/deepseek/deepseek_v2_bridge.py

@@ -12,41 +12,77 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import torch
+from functools import partial
+
+from megatron.core.models.gpt.gpt_layer_specs import get_gpt_decoder_block_spec
 from megatron.core.models.gpt.gpt_model import GPTModel
 
 from megatron.bridge.models.conversion.mapping_registry import MegatronMappingRegistry
 from megatron.bridge.models.conversion.model_bridge import MegatronModelBridge
-from megatron.bridge.models.deepseek.common import get_common_configs, get_common_mapping_list
-from megatron.bridge.models.deepseek.deepseek_provider import DeepSeekV2ModelProvider
+from megatron.bridge.models.deepseek.common import get_common_mapping_list
 from megatron.bridge.models.hf_pretrained.causal_lm import PreTrainedCausalLM
+from megatron.bridge.models.mla_provider import MLAModelProvider
 
 
-@MegatronModelBridge.register_bridge(source="DeepseekV2ForCausalLM", target=GPTModel)
-class DeepSeekV2Bridge(MegatronModelBridge):
-    """
-    Megatron Bridge for DeepSeek-V2.
+try:
+    import transformer_engine  # noqa: F401
 
-    As a user you would not use this bridge directly, but through `AutoBridge`.
+    HAVE_TE = True
+except (ImportError, ModuleNotFoundError):
+    HAVE_TE = False
 
-    Example:
-        >>> from megatron.bridge import AutoBridge
-        >>> bridge = AutoBridge.from_hf_pretrained("deepseek-ai/DeepSeek-V2", trust_remote_code=True)
-        >>> provider = bridge.to_megatron_provider()
-    """
 
-    def provider_bridge(self, hf_pretrained: PreTrainedCausalLM) -> DeepSeekV2ModelProvider:
+@MegatronModelBridge.register_bridge(
+    source="DeepseekV2ForCausalLM",
+    target=GPTModel,
+    provider=MLAModelProvider,
+    model_type="deepseek_v2",
+)
+class DeepSeekV2Bridge(MegatronModelBridge):
+    """Megatron Bridge for DeepSeek-V2."""
+
+    def provider_bridge(self, hf_pretrained: PreTrainedCausalLM) -> MLAModelProvider:
+        provider = super().provider_bridge(hf_pretrained)
         hf_config = hf_pretrained.config
-        configs = get_common_configs(hf_pretrained)
 
-        configs["fp16"] = self.dtype_from_hf(hf_config, default=torch.float32) == torch.float16
-        configs["bf16"] = self.dtype_from_hf(hf_config, default=torch.float32) == torch.bfloat16
-        configs["params_dtype"] = self.dtype_from_hf(hf_config, default=torch.float32)
+        provider.transformer_layer_spec = partial(get_gpt_decoder_block_spec, use_transformer_engine=HAVE_TE)
+        provider.normalization = "RMSNorm"
+        provider.gated_linear_unit = True
+        provider.position_embedding_type = "rope"
+        provider.add_bias_linear = False
+        provider.share_embeddings_and_output_weights = False
+        provider.qk_layernorm = True
+        provider.multi_latent_attention = True
+
+        provider.moe_grouped_gemm = True
+        provider.moe_router_pre_softmax = True
+        provider.moe_token_dispatcher_type = "alltoall"
+        provider.moe_router_load_balancing_type = "seq_aux_loss"
+        provider.moe_shared_expert_overlap = True
+        provider.moe_router_dtype = "fp32"
+        provider.moe_permute_fusion = True
+
+        provider.apply_rope_fusion = False
+        provider.bias_activation_fusion = True
+        provider.bias_dropout_fusion = True
+        provider.cross_entropy_fusion_impl = "te"
+        provider.cross_entropy_loss_fusion = True
+        provider.masked_softmax_fusion = True
+        provider.persist_layer_norm = True
+        provider.async_tensor_model_parallel_allreduce = True
+        provider.gradient_accumulation_fusion = True
+
+        provider.hidden_dropout = 0.0
+        provider.attention_softmax_in_fp32 = False
+
+        provider.make_vocab_size_divisible_by = 3200
+        provider.seq_length = 4096
 
-        configs["make_vocab_size_divisible_by"] = 3200
-        configs["moe_aux_loss_coeff"] = hf_config.aux_loss_alpha
+        provider.moe_layer_freq = [0] * hf_config.first_k_dense_replace + [1] * (
+            hf_config.num_hidden_layers - hf_config.first_k_dense_replace
+        )
+        provider.moe_shared_expert_intermediate_size = hf_config.moe_intermediate_size * hf_config.n_shared_experts
 
-        provider = DeepSeekV2ModelProvider(**configs)
         return provider
 
     def mapping_registry(self) -> MegatronMappingRegistry: