bridge

Megatron Bridge - A component of the Megatron ecosystem.

Subpackages

• bridge.peft
  • bridge.peft.dora_layers
  • bridge.peft.lora
  • bridge.peft.dora
  • bridge.peft.adapter_wrapper
  • bridge.peft.walk_utils
  • bridge.peft.base
  • bridge.peft.module_matcher
  • bridge.peft.lora_layers
  • bridge.peft.canonical_lora
  • bridge.peft.utils
• bridge.data
  • bridge.data.datasets
    • bridge.data.datasets.sft
    • bridge.data.datasets.packed_sequence
    • bridge.data.datasets.packing_utils
    • bridge.data.datasets.utils
  • bridge.data.builders
    • bridge.data.builders.hf_dataset
    • bridge.data.builders.finetuning_dataset
  • bridge.data.hf_processors
    • bridge.data.hf_processors.squad
  • bridge.data.loaders
  • bridge.data.samplers
  • bridge.data.utils
• bridge.models
  • bridge.models.conversion
    • bridge.models.conversion.auto_bridge
    • bridge.models.conversion.param_mapping
    • bridge.models.conversion.mapping_registry
    • bridge.models.conversion.utils
    • bridge.models.conversion.model_bridge
  • bridge.models.qwen
    • bridge.models.qwen.qwen3_bridge
    • bridge.models.qwen.qwen_provider
    • bridge.models.qwen.qwen3_moe_bridge
    • bridge.models.qwen.qwen2_bridge
  • bridge.models.llama
    • bridge.models.llama.llama_bridge
    • bridge.models.llama.llama_provider
    • bridge.models.llama.llama4_utils
  • bridge.models.mamba
    • bridge.models.mamba.nemotron_h_provider
    • bridge.models.mamba.mamba_provider
  • bridge.models.deepseek
    • bridge.models.deepseek.deepseek_v3_bridge
    • bridge.models.deepseek.common
    • bridge.models.deepseek.deepseek_v2_bridge
    • bridge.models.deepseek.deepseek_provider
  • bridge.models.decorators
    • bridge.models.decorators.torchrun
    • bridge.models.decorators.dispatch
  • bridge.models.hf_pretrained
    • bridge.models.hf_pretrained.causal_lm
    • bridge.models.hf_pretrained.state
    • bridge.models.hf_pretrained.base
    • bridge.models.hf_pretrained.vlm
  • bridge.models.gpt_full_te_layer_autocast_spec
  • bridge.models.model_provider
  • bridge.models.config
  • bridge.models.gpt_provider
  • bridge.models.transformer_config
  • bridge.models.t5_provider
• bridge.recipes
  • bridge.recipes.qwen
    • bridge.recipes.qwen.qwen25_72b
    • bridge.recipes.qwen.qwen3_8b
    • bridge.recipes.qwen.qwen25_32b
    • bridge.recipes.qwen.qwen3_30b_a3b
    • bridge.recipes.qwen.qwen3_235b_a22b
    • bridge.recipes.qwen.qwen3_32b
    • bridge.recipes.qwen.qwen2_1p5b
    • bridge.recipes.qwen.qwen25_500m
    • bridge.recipes.qwen.qwen2_7b
    • bridge.recipes.qwen.qwen3_14b
    • bridge.recipes.qwen.qwen25_14b
    • bridge.recipes.qwen.qwen3_600m
    • bridge.recipes.qwen.qwen25_1p5b
    • bridge.recipes.qwen.qwen3_4b
    • bridge.recipes.qwen.qwen2_72b
    • bridge.recipes.qwen.qwen25_7b
    • bridge.recipes.qwen.qwen3_1p7b
    • bridge.recipes.qwen.qwen2_500m
  • bridge.recipes.gpt
    • bridge.recipes.gpt.gpt3_175b
  • bridge.recipes.utils
    • bridge.recipes.utils.tokenizer_utils
    • bridge.recipes.utils.nemo_run_utils
    • bridge.recipes.utils.dataset_utils
    • bridge.recipes.utils.optimizer_utils
  • bridge.recipes.deepseek
    • bridge.recipes.deepseek.deepseek_v2_lite
    • bridge.recipes.deepseek.deepseek_v3
    • bridge.recipes.deepseek.deepseek_v2
  • bridge.recipes.run_plugins
• bridge.utils
  • bridge.utils.vocab_utils
  • bridge.utils.instantiate_utils
  • bridge.utils.common_utils
  • bridge.utils.fusions
  • bridge.utils.yaml_utils
  • bridge.utils.import_utils
  • bridge.utils.decorators
• bridge.training
  • bridge.training.tokenizers
    • bridge.training.tokenizers.gpt2_tokenization
    • bridge.training.tokenizers.tokenizer
    • bridge.training.tokenizers.config
    • bridge.training.tokenizers.bert_tokenization
    • bridge.training.tokenizers.multimodal_tokenizer
  • bridge.training.mlm_compat
    • bridge.training.mlm_compat.model
    • bridge.training.mlm_compat.activations
    • bridge.training.mlm_compat.arguments
  • bridge.training.utils
    • bridge.training.utils.omegaconf_utils
    • bridge.training.utils.log_utils
    • bridge.training.utils.moe_token_drop
    • bridge.training.utils.train_utils
    • bridge.training.utils.flop_utils
    • bridge.training.utils.theoretical_memory_utils
    • bridge.training.utils.sig_utils
    • bridge.training.utils.wandb_utils
    • bridge.training.utils.config_utils
    • bridge.training.utils.checkpoint_utils
  • bridge.training.model_load_save
  • bridge.training.fault_tolerance
  • bridge.training.pretrain
  • bridge.training.optim
  • bridge.training.state
  • bridge.training.train
  • bridge.training.nvrx_straggler
  • bridge.training.losses
  • bridge.training.eval
  • bridge.training.setup
  • bridge.training.initialize
  • bridge.training.checkpointing
  • bridge.training.inprocess_restart
  • bridge.training.config
  • bridge.training.comm_overlap
  • bridge.training.mixed_precision
  • bridge.training.finetune
  • bridge.training.deepep
  • bridge.training.gpt_step

Submodules

• bridge.package_info

Package Contents

Data

__all__

API

bridge.__all__

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