bridge.models.sarvam.sarvam_mla_bridge#

Module Contents#

Classes#

SarvamMLABridge

Megatron Hub Bridge for Sarvam MLA Causal LM.

API#

class bridge.models.sarvam.sarvam_mla_bridge.SarvamMLABridge#

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

Megatron Hub Bridge for Sarvam MLA Causal LM.

This bridge handles the conversion between HuggingFace SarvamMLAForCausalLM and Megatron-Core GPTModel formats. Sarvam MLA models use multi-latent attention architecture.

provider_bridge(
hf_pretrained: megatron.bridge.models.hf_pretrained.causal_lm.PreTrainedCausalLM,
) → megatron.bridge.models.sarvam.sarvam_provider.SarvamMLAModelProvider#
mapping_registry() → megatron.bridge.models.conversion.mapping_registry.MegatronMappingRegistry#