bridge

Megatron Bridge - A component of the Megatron ecosystem.

Subpackages

• bridge.recipes
  • bridge.recipes.kimi_vl
    • bridge.recipes.kimi_vl.kimi_k25_vl
  • bridge.recipes.llama
    • bridge.recipes.llama.llama3
    • bridge.recipes.llama.llama2
  • bridge.recipes.ministral3
    • bridge.recipes.ministral3.ministral3
  • bridge.recipes.utils
    • bridge.recipes.utils.optimizer_utils
    • bridge.recipes.utils.tokenizer_utils
    • bridge.recipes.utils.determinism_utils
    • bridge.recipes.utils.finetune_utils
    • bridge.recipes.utils.dataset_utils
  • bridge.recipes.qwen_omni
    • bridge.recipes.qwen_omni.qwen3_omni
  • bridge.recipes.gemma
    • bridge.recipes.gemma.gemma3
    • bridge.recipes.gemma.gemma2
  • bridge.recipes.nemotronh
    • bridge.recipes.nemotronh.nemotron_3_nano
    • bridge.recipes.nemotronh.nemotron_3_super
    • bridge.recipes.nemotronh.nemotronh
    • bridge.recipes.nemotronh.nemotron_nano_v2
  • bridge.recipes.flux
  • bridge.recipes.stepfun
    • bridge.recipes.stepfun.step37
    • bridge.recipes.stepfun.step35
  • bridge.recipes.glm
    • bridge.recipes.glm.glm45
  • bridge.recipes.gemma4_vl
    • bridge.recipes.gemma4_vl.gemma4_vl
  • bridge.recipes.glm_vl
    • bridge.recipes.glm_vl.glm_45v
  • bridge.recipes.qwen_vl
    • bridge.recipes.qwen_vl.qwen3_vl
    • bridge.recipes.qwen_vl.qwen25_vl_dataset
    • bridge.recipes.qwen_vl.qwen35_vl
    • bridge.recipes.qwen_vl.qwen25_vl
  • bridge.recipes.gpt
    • bridge.recipes.gpt.gpt3_175b
    • bridge.recipes.gpt.vanilla_gpt
  • bridge.recipes.wan
  • bridge.recipes.gemma3_vl
    • bridge.recipes.gemma3_vl.gemma3_vl
  • bridge.recipes.qwen
    • bridge.recipes.qwen.qwen3_next
    • bridge.recipes.qwen.qwen3_moe
    • bridge.recipes.qwen.qwen2
    • bridge.recipes.qwen.qwen3
  • bridge.recipes.moonlight
    • bridge.recipes.moonlight.moonlight_16b
  • bridge.recipes.gpt_oss
    • bridge.recipes.gpt_oss.gpt_oss
  • bridge.recipes.deepseek
    • bridge.recipes.deepseek.deepseek_v2
    • bridge.recipes.deepseek.deepseek_v3
    • bridge.recipes.deepseek.deepseek_v4
  • bridge.recipes.qwen2_audio
    • bridge.recipes.qwen2_audio.qwen2_audio
  • bridge.recipes.nemotron_vl
    • bridge.recipes.nemotron_vl.nemotron_nano_v2_vl
  • bridge.recipes.olmoe
    • bridge.recipes.olmoe.olmoe_7b
  • bridge.recipes.nemotron_omni
    • bridge.recipes.nemotron_omni.nemotron_omni
  • bridge.recipes.common
  • bridge.recipes.run_plugins
• bridge.utils
  • bridge.utils.decorators
  • bridge.utils.yaml_utils
  • bridge.utils.safe_url
  • bridge.utils.cuda_graph
  • bridge.utils.safe_pickle
  • bridge.utils.activation_map
  • bridge.utils.import_utils
  • bridge.utils.instantiate_utils
  • bridge.utils.fusions
  • bridge.utils.common_utils
  • bridge.utils.slurm_utils
  • bridge.utils.vocab_utils
• bridge.training
  • bridge.training.utils
    • bridge.training.utils.omegaconf_utils
    • bridge.training.utils.log_utils
    • bridge.training.utils.train_utils
    • bridge.training.utils.wandb_utils
    • bridge.training.utils.theoretical_memory_utils
    • bridge.training.utils.sig_utils
    • bridge.training.utils.moe_token_drop
    • bridge.training.utils.flop_utils
    • bridge.training.utils.comet_utils
    • bridge.training.utils.config_utils
    • bridge.training.utils.padding_utils
    • bridge.training.utils.packed_seq_utils
    • bridge.training.utils.checkpoint_utils
    • bridge.training.utils.pg_utils
    • bridge.training.utils.mlflow_utils
    • bridge.training.utils.batch_utils
    • bridge.training.utils.visual_inputs
  • bridge.training.mlm_compat
    • bridge.training.mlm_compat.arguments
    • bridge.training.mlm_compat.activations
    • bridge.training.mlm_compat.model
  • bridge.training.tokenizers
    • bridge.training.tokenizers.tokenizer
    • bridge.training.tokenizers.config
  • bridge.training.post_training
    • bridge.training.post_training.distillation
    • bridge.training.post_training.checkpointing
  • bridge.training.finetune
  • bridge.training.flex_dispatcher_backend
  • bridge.training.gpt_step
  • bridge.training.inprocess_restart
  • bridge.training.state
  • bridge.training.train
  • bridge.training.distill
  • bridge.training.train_megatron_mimo
  • bridge.training.vlm_step
  • bridge.training.callbacks
  • bridge.training.megatron_mimo_step
  • bridge.training.tensor_inspect
  • bridge.training.megatron_mimo_parallel_utils
  • bridge.training.nvrx_straggler
  • bridge.training.model_load_save
  • bridge.training.nemotron_omni_step
  • bridge.training.comm_overlap
  • bridge.training.eval
  • bridge.training.optim
  • bridge.training.setup_megatron_mimo
  • bridge.training.forward_step_func_types
  • bridge.training.config
  • bridge.training.fault_tolerance
  • bridge.training.checkpointing
  • bridge.training.audio_lm_step
  • bridge.training.setup
  • bridge.training.pretrain_megatron_mimo
  • bridge.training.profiling
  • bridge.training.initialize
  • bridge.training.losses
  • bridge.training.llava_step
  • bridge.training.mixed_precision
  • bridge.training.pretrain
  • bridge.training.eval_context_parallel_rebinding
• bridge.peft
  • bridge.peft.canonical_lora
  • bridge.peft.module_matcher
  • bridge.peft.lora_layers
  • bridge.peft.dora
  • bridge.peft.utils
  • bridge.peft.walk_utils
  • bridge.peft.dora_layers
  • bridge.peft.base
  • bridge.peft.lora
  • bridge.peft.adapter_wrapper
  • bridge.peft.recompute
• bridge.diffusion
  • bridge.diffusion.recipes
    • bridge.diffusion.recipes.flux
    • bridge.diffusion.recipes.wan
    • bridge.diffusion.recipes.nemotron_labs_diffusion
  • bridge.diffusion.conversion
    • bridge.diffusion.conversion.flux
    • bridge.diffusion.conversion.wan
    • bridge.diffusion.conversion.nemotron_labs_diffusion
  • bridge.diffusion.base
  • bridge.diffusion.common
    • bridge.diffusion.common.flow_matching
    • bridge.diffusion.common.dllm
  • bridge.diffusion.models
    • bridge.diffusion.models.flux
    • bridge.diffusion.models.wan
    • bridge.diffusion.models.common
    • bridge.diffusion.models.nemotron_labs_diffusion
  • bridge.diffusion.data
    • bridge.diffusion.data.flux
    • bridge.diffusion.data.wan
    • bridge.diffusion.data.common
• bridge.models
  • bridge.models.nemotron
    • bridge.models.nemotron.nemotron_bridge
  • bridge.models.kimi_vl
    • bridge.models.kimi_vl.kimi_k25_vl_bridge
    • bridge.models.kimi_vl.utils
    • bridge.models.kimi_vl.kimi_k25_vl_provider
    • bridge.models.kimi_vl.modeling_kimi_k25_vl
  • bridge.models.llama
    • bridge.models.llama.llama_bridge
    • bridge.models.llama.llama4_utils
  • bridge.models.kimi
    • bridge.models.kimi.kimi_bridge
  • bridge.models.qwen_audio
    • bridge.models.qwen_audio.qwen2_audio_bridge
    • bridge.models.qwen_audio.modeling_qwen2_audio
    • bridge.models.qwen_audio.qwen2_audio_provider
  • bridge.models.ministral3
    • bridge.models.ministral3.modeling_ministral3
    • bridge.models.ministral3.ministral3_provider
    • bridge.models.ministral3.ministral3_bridge
  • bridge.models.qwen_omni
    • bridge.models.qwen_omni.modeling_qwen25_omni
    • bridge.models.qwen_omni.modeling_qwen3_omni
    • bridge.models.qwen_omni.qwen3_omni_step
    • bridge.models.qwen_omni.qwen25_omni_provider
    • bridge.models.qwen_omni.qwen3_omni_bridge
    • bridge.models.qwen_omni.qwen3_omni_provider
    • bridge.models.qwen_omni.qwen25_omni_bridge
  • bridge.models.sarvam
    • bridge.models.sarvam.common
    • bridge.models.sarvam.sarvam_mla_bridge
    • bridge.models.sarvam.sarvam_moe_bridge
    • bridge.models.sarvam.sarvam_provider
  • bridge.models.gemma
    • bridge.models.gemma.gemma3_provider
    • bridge.models.gemma.gemma4_provider
    • bridge.models.gemma.modules
    • bridge.models.gemma.gemma_bridge
    • bridge.models.gemma.gemma_provider
    • bridge.models.gemma.gemma2_provider
    • bridge.models.gemma.gemma4_bridge
    • bridge.models.gemma.gemma2_bridge
    • bridge.models.gemma.gemma3_bridge
  • bridge.models.nemotronh
    • bridge.models.nemotronh.nemotron_h_bridge
  • bridge.models.glm_moe_dsa
    • bridge.models.glm_moe_dsa.glm5_bridge
    • bridge.models.glm_moe_dsa.glm5_provider
  • bridge.models.mimo_v2_flash
    • bridge.models.mimo_v2_flash.modeling_mimo_v2_flash
    • bridge.models.mimo_v2_flash.mimo_v2_flash_bridge
    • bridge.models.mimo_v2_flash.mimo_v2_flash_provider
  • bridge.models.ernie_vl
    • bridge.models.ernie_vl.modeling_ernie45_vl
    • bridge.models.ernie_vl.ernie45_vl_bridge
    • bridge.models.ernie_vl.ernie45_vl_provider
  • bridge.models.stepfun
    • bridge.models.stepfun.modelling_step37
    • bridge.models.stepfun.step37_bridge
    • bridge.models.stepfun.configuration_step37
    • bridge.models.stepfun.step35_bridge
    • bridge.models.stepfun.step37_flickr8k_step
    • bridge.models.stepfun.configuration_step35
    • bridge.models.stepfun.step35_provider
    • bridge.models.stepfun.step37_provider
  • bridge.models.mamba
    • bridge.models.mamba.mamba_provider
    • bridge.models.mamba.mamba_builder
  • bridge.models.conversion
    • bridge.models.conversion.quantization_utils
    • bridge.models.conversion.mapping_registry
    • bridge.models.conversion.model_bridge
    • bridge.models.conversion.auto_bridge
    • bridge.models.conversion.peft_bridge
    • bridge.models.conversion.utils
    • bridge.models.conversion.transformers_compat
    • bridge.models.conversion.quant_mapping
    • bridge.models.conversion.quant_bridge
    • bridge.models.conversion.param_mapping
  • bridge.models.glm
    • bridge.models.glm.glm45_bridge
    • bridge.models.glm.glm47_flash_bridge
  • bridge.models.minimax_m2
    • bridge.models.minimax_m2.minimax_m2_bridge
    • bridge.models.minimax_m2.minimax_m2_provider
  • bridge.models.qwen3_asr
    • bridge.models.qwen3_asr.modeling_qwen3_asr
    • bridge.models.qwen3_asr.hf_qwen3_asr
    • bridge.models.qwen3_asr.qwen3_asr_provider
    • bridge.models.qwen3_asr.qwen3_asr_bridge
  • bridge.models.glm_vl
    • bridge.models.glm_vl.glm_45v_provider
    • bridge.models.glm_vl.modeling_glm_45v
    • bridge.models.glm_vl.glm_45v_bridge
  • bridge.models.megatron_mimo
    • bridge.models.megatron_mimo.conversion
    • bridge.models.megatron_mimo.megatron_mimo_provider
    • bridge.models.megatron_mimo.megatron_mimo_builder
    • bridge.models.megatron_mimo.megatron_mimo_ddp
    • bridge.models.megatron_mimo.build_model
    • bridge.models.megatron_mimo.megatron_mimo_config
    • bridge.models.megatron_mimo.llava_provider
  • bridge.models.llama_nemotron
    • bridge.models.llama_nemotron.llama_nemotron_provider
    • bridge.models.llama_nemotron.llama_nemotron_bridge
  • bridge.models.hf_pretrained
    • bridge.models.hf_pretrained.state
    • bridge.models.hf_pretrained.utils
    • bridge.models.hf_pretrained.base
    • bridge.models.hf_pretrained.safe_config_loader
    • bridge.models.hf_pretrained.vlm
    • bridge.models.hf_pretrained.causal_lm
  • bridge.models.gemma_vl
    • bridge.models.gemma_vl.gemma3_vl_bridge
    • bridge.models.gemma_vl.gemma3_vl_provider
    • bridge.models.gemma_vl.gemma4_vl_provider
    • bridge.models.gemma_vl.modeling_gemma4_vl
    • bridge.models.gemma_vl.modeling_gemma3_vl
    • bridge.models.gemma_vl.gemma4_vl_bridge
  • bridge.models.qwen_vl
    • bridge.models.qwen_vl.modelling_qwen3_vl
    • bridge.models.qwen_vl.qwen3_vl_provider
    • bridge.models.qwen_vl.qwen35_vl_bridge
    • bridge.models.qwen_vl.qwen3_vl_bridge
    • bridge.models.qwen_vl.qwen25_vl_bridge
    • bridge.models.qwen_vl.qwen25_vl_provider
    • bridge.models.qwen_vl.modeling_qwen25_vl
    • bridge.models.qwen_vl.qwen35_vl_provider
    • bridge.models.qwen_vl.qwen3_vl_step
  • bridge.models.ernie
    • bridge.models.ernie.ernie_45_bridge
  • bridge.models.decorators
    • bridge.models.decorators.dispatch
    • bridge.models.decorators.torchrun
  • bridge.models.gpt
    • bridge.models.gpt.gpt_builder
  • bridge.models.qwen
    • bridge.models.qwen.qwen3_next_bridge
    • bridge.models.qwen.qwen2_bridge
    • bridge.models.qwen.qwen3_bridge
    • bridge.models.qwen.qwen3_moe_bridge
  • bridge.models.mistral
    • bridge.models.mistral.mistral_provider
    • bridge.models.mistral.mistral_bridge
  • bridge.models.gpt_oss
    • bridge.models.gpt_oss.gpt_oss_bridge
  • bridge.models.common
    • bridge.models.common.unimodal
    • bridge.models.common.base
  • bridge.models.deepseek
    • bridge.models.deepseek.common
    • bridge.models.deepseek.deepseek_v3_bridge
    • bridge.models.deepseek.deepseek_v2_bridge
    • bridge.models.deepseek.deepseek_v4_bridge
  • bridge.models.mimo
    • bridge.models.mimo.mimo_bridge
  • bridge.models.bailing
    • bridge.models.bailing.configuration_bailing_moe_v2
    • bridge.models.bailing.bailing_moe2_bridge
    • bridge.models.bailing.modeling_bailing_moe_v2
  • bridge.models.nemotron_vl
    • bridge.models.nemotron_vl.nemotron_vl_utils
    • bridge.models.nemotron_vl.nemotron_vl_provider
    • bridge.models.nemotron_vl.modeling_nemotron_vl
    • bridge.models.nemotron_vl.nemotron_vl_bridge
  • bridge.models.olmoe
    • bridge.models.olmoe.olmoe_bridge
    • bridge.models.olmoe.olmoe_provider
  • bridge.models.nemotron_omni
    • bridge.models.nemotron_omni.nemotron_omni_bridge
    • bridge.models.nemotron_omni.modeling_nemotron_omni
    • bridge.models.nemotron_omni.nemotron_omni_utils
    • bridge.models.nemotron_omni.nemotron_omni_sound
    • bridge.models.nemotron_omni.nemotron_omni_provider
  • bridge.models.falcon_h1
    • bridge.models.falcon_h1.falconh1_provider
    • bridge.models.falcon_h1.falconh1_bridge
  • bridge.models.t5_provider
  • bridge.models.transformer_config
  • bridge.models.mla_provider
  • bridge.models.config
  • bridge.models.distillation_provider
  • bridge.models.gpt_full_te_layer_autocast_spec
  • bridge.models.model_provider
  • bridge.models.gpt_provider
• bridge.data
  • bridge.data.vlm_datasets
    • bridge.data.vlm_datasets.step37_flickr8k
    • bridge.data.vlm_datasets.hf_provider
    • bridge.data.vlm_datasets.token_utils
    • bridge.data.vlm_datasets.collate
    • bridge.data.vlm_datasets.preloaded_provider
    • bridge.data.vlm_datasets.hf_dataset_makers
    • bridge.data.vlm_datasets.conversation_dataset
    • bridge.data.vlm_datasets.mock_provider
  • bridge.data.hf_processors
    • bridge.data.hf_processors.openmathinstruct2
    • bridge.data.hf_processors.squad
    • bridge.data.hf_processors.gsm8k
  • bridge.data.builders
    • bridge.data.builders.hf_dataset
    • bridge.data.builders.finetuning_dataset
  • bridge.data.megatron_mimo
    • bridge.data.megatron_mimo.dataset
    • bridge.data.megatron_mimo.hf_provider
    • bridge.data.megatron_mimo.collate
    • bridge.data.megatron_mimo.dp_utils
    • bridge.data.megatron_mimo.loaders
    • bridge.data.megatron_mimo.base_provider
    • bridge.data.megatron_mimo.mock_provider
  • bridge.data.energon
    • bridge.data.energon.energon_provider
    • bridge.data.energon.base_energon_datamodule
    • bridge.data.energon.metadata
    • bridge.data.energon.nemotron_omni_task_encoder
    • bridge.data.energon.task_encoder_utils
    • bridge.data.energon.hf_encoder_task_encoder
  • bridge.data.datasets
    • bridge.data.datasets.sft
    • bridge.data.datasets.packing_utils
    • bridge.data.datasets.packed_sequence
    • bridge.data.datasets.utils
    • bridge.data.datasets.packed_parquet
    • bridge.data.datasets.fim_dataset
  • bridge.data.utils
  • bridge.data.loaders
  • bridge.data.samplers
  • bridge.data.finetuning
  • bridge.data.iterator_utils

Submodules

• bridge.package_info

Package Contents

Data

__all__

API

bridge.__all__

[’version’, ‘package_name’, ‘contact_names’, ‘contact_emails’, ‘homepage’, ‘__re…