`bridge.models.bailing.bailing_moe2_bridge`#

Megatron Bridge for Ling MoE2 Model.

This module provides the bridge implementation for converting between HuggingFace Bailing MoE2 models and Megatron-Core format.

Supported models:

Module Contents#

Megatron Bridge for Ling MoE V2 Model

class bridge.models.bailing.bailing_moe2_bridge.BailingMoeV2Bridge#

Bases: megatron.bridge.models.conversion.model_bridge.MegatronModelBridge

Megatron Bridge for Ling MoE V2 Model

.. rubric:: Example

from megatron.bridge import AutoBridge bridge = AutoBridge.from_hf_pretrained(“inclusionAI/Ling-mini-2.0”) model_config = bridge.get_model_config()

hf_config_to_model_config_kwargs( hf_config: transformers.PretrainedConfig, ) → dict[str, Any]#

Convert a Hugging Face Ling MoE V2 config to builder config kwargs.

The builder’s default MoE layer spec already uses the GPT decoder block spec, which honors the dense-first moe_layer_freq.

hf_config_to_provider_kwargs( hf_config: transformers.PretrainedConfig, ) → dict[str, Any]#: Adapt the canonical builder mapping to the deprecated provider path.

mapping_registry() → megatron.bridge.models.conversion.mapping_registry.MegatronMappingRegistry#