bridge.models.qwen.qwen3_next_bridge#
Module Contents#
Classes#
Megatron Bridge for Qwen3-Next Causal LM. |
API#
- class bridge.models.qwen.qwen3_next_bridge.Qwen3NextBridge#
Bases:
megatron.bridge.models.conversion.model_bridge.MegatronModelBridgeMegatron Bridge for Qwen3-Next Causal LM.
This bridge handles the conversion between HuggingFace Qwen3NextForCausalLM and Megatron-Core GPTModel formats. Qwen3-Next uses a hybrid architecture combining gated delta net linear attention with standard softmax attention, mixture of experts with shared experts, and zero-centered RMSNorm.
.. rubric:: Example
from megatron.bridge import AutoBridge bridge = AutoBridge.from_hf_pretrained(“Qwen/Qwen3-Next-80B-A3B-Instruct”) provider = bridge.to_megatron_provider()
- provider_bridge(hf_pretrained)#
Convert HuggingFace Qwen3-Next config to GPTModelProvider.
- mapping_registry() megatron.bridge.models.conversion.mapping_registry.MegatronMappingRegistry#