core

Subpackages

• core.dist_checkpointing
  • core.dist_checkpointing.strategies
    • core.dist_checkpointing.strategies.fully_parallel
    • core.dist_checkpointing.strategies.checkpointable
    • core.dist_checkpointing.strategies.zarr
    • core.dist_checkpointing.strategies.cached_metadata_filesystem_reader
    • core.dist_checkpointing.strategies.common
    • core.dist_checkpointing.strategies.torch
    • core.dist_checkpointing.strategies.async_utils
    • core.dist_checkpointing.strategies.filesystem_async
    • core.dist_checkpointing.strategies.tensorstore
    • core.dist_checkpointing.strategies.two_stage
    • core.dist_checkpointing.strategies.base
    • core.dist_checkpointing.strategies.state_dict_saver
  • core.dist_checkpointing.optimizer
  • core.dist_checkpointing.mapping
  • core.dist_checkpointing.dict_utils
  • core.dist_checkpointing.serialization
  • core.dist_checkpointing.validation
  • core.dist_checkpointing.core
  • core.dist_checkpointing.state_dict_utils
  • core.dist_checkpointing.tensor_aware_state_dict
  • core.dist_checkpointing.exchange_utils
  • core.dist_checkpointing.utils
• core.export
  • core.export.trtllm
    • core.export.trtllm.engine_builder
    • core.export.trtllm.trtllm_weights_converter
    • core.export.trtllm.model_to_trllm_mapping
    • core.export.trtllm.trt_model_config
    • core.export.trtllm.trtllm_helper
    • core.export.trtllm.trt_model_type
    • core.export.trtllm.trtllm_layers
  • core.export.data_type
  • core.export.export_config
  • core.export.model_type
• core.extensions
  • core.extensions.kitchen
  • core.extensions.transformer_engine
  • core.extensions.transformer_engine_spec_provider
• core.tokenizers
  • core.tokenizers.text
    • core.tokenizers.text.libraries
    • core.tokenizers.text.models
    • core.tokenizers.text.text_tokenizer
  • core.tokenizers.base_tokenizer
  • core.tokenizers.megatron_tokenizer
• core.models
  • core.models.retro
    • core.models.retro.encoder_attention
    • core.models.retro.model
    • core.models.retro.decoder_spec
    • core.models.retro.encoder_spec
    • core.models.retro.base_attention
    • core.models.retro.config
    • core.models.retro.decoder_attention
    • core.models.retro.utils
  • core.models.vision
    • core.models.vision.clip_vit_model
    • core.models.vision.multimodal_projector
    • core.models.vision.radio
    • core.models.vision.vit_layer_specs
  • core.models.gpt
    • core.models.gpt.moe_module_specs
    • core.models.gpt.gpt_model
    • core.models.gpt.fine_grained_callables
    • core.models.gpt.gpt_layer_specs
  • core.models.T5
    • core.models.T5.t5_spec
    • core.models.T5.t5_model
  • core.models.multimodal
    • core.models.multimodal.llava_model
    • core.models.multimodal.llava_spec
    • core.models.multimodal.context_parallel
  • core.models.mamba
    • core.models.mamba.mamba_layer_specs
    • core.models.mamba.mamba_model
  • core.models.common
    • core.models.common.embeddings
    • core.models.common.vision_module
    • core.models.common.language_module
    • core.models.common.model_chunk_schedule_plan
  • core.models.huggingface
    • core.models.huggingface.clip_model
    • core.models.huggingface.module
    • core.models.huggingface.qwen_model
  • core.models.mimo
    • core.models.mimo.config
    • core.models.mimo.model
  • core.models.bert
    • core.models.bert.bert_layer_specs
    • core.models.bert.bert_model
    • core.models.bert.pooler
    • core.models.bert.bert_lm_head
  • core.models.backends
• core.tensor_parallel
  • core.tensor_parallel.data
  • core.tensor_parallel.layers
  • core.tensor_parallel.inference_layers
  • core.tensor_parallel.mappings
  • core.tensor_parallel.random
  • core.tensor_parallel.cross_entropy
  • core.tensor_parallel.utils
• core.distributed
  • core.distributed.fsdp
    • core.distributed.fsdp.src
    • core.distributed.fsdp.mcore_fsdp_adapter
  • core.distributed.param_and_grad_buffer
  • core.distributed.torch_fully_sharded_data_parallel
  • core.distributed.torch_fully_sharded_data_parallel_config
  • core.distributed.finalize_model_grads
  • core.distributed.distributed_data_parallel_config
  • core.distributed.reduce_scatter_with_fp32_accumulation
  • core.distributed.data_parallel_base
  • core.distributed.distributed_data_parallel
• core.datasets
  • core.datasets.retro
    • core.datasets.retro.config
    • core.datasets.retro.index
    • core.datasets.retro.query
    • core.datasets.retro.db
    • core.datasets.retro.external_libs
    • core.datasets.retro.utils
  • core.datasets.blended_megatron_dataset_builder
  • core.datasets.multimodal_dataset
  • core.datasets.t5_dataset
  • core.datasets.gpt_dataset
  • core.datasets.helpers
  • core.datasets.object_storage_utils
  • core.datasets.bert_dataset
  • core.datasets.utils_s3
  • core.datasets.megatron_tokenizer
  • core.datasets.megatron_dataset
  • core.datasets.blended_dataset
  • core.datasets.masked_dataset
  • core.datasets.indexed_dataset
  • core.datasets.utils
  • core.datasets.blended_megatron_dataset_config
• core.quantization
  • core.quantization.quant_config
  • core.quantization.utils
• core.ssm
  • core.ssm.mamba_block
  • core.ssm.mamba_context_parallel
  • core.ssm.triton_cache_manager
  • core.ssm.mamba_mixer
  • core.ssm.mlp_layer
  • core.ssm.mamba_layer
  • core.ssm.mamba_hybrid_layer_allocation
• core.transformer
  • core.transformer.moe
    • core.transformer.moe.shared_experts
    • core.transformer.moe.router
    • core.transformer.moe.moe_utils
    • core.transformer.moe.upcycling_utils
    • core.transformer.moe.moe_layer
    • core.transformer.moe.grouped_gemm_util
    • core.transformer.moe.experts
    • core.transformer.moe.fused_a2a
    • core.transformer.moe.token_dispatcher
  • core.transformer.custom_layers
    • core.transformer.custom_layers.transformer_engine
    • core.transformer.custom_layers.batch_invariant_kernels
  • core.transformer.torch_layer_norm
  • core.transformer.transformer_layer
  • core.transformer.multi_token_prediction
  • core.transformer.module
  • core.transformer.mlp
  • core.transformer.transformer_block
  • core.transformer.identity_op
  • core.transformer.multi_latent_attention
  • core.transformer.pipeline_parallel_layer_layout
  • core.transformer.attention
  • core.transformer.fsdp_dtensor_checkpoint
  • core.transformer.enums
  • core.transformer.spec_utils
  • core.transformer.transformer_config
  • core.transformer.cuda_graphs
  • core.transformer.torch_norm
  • core.transformer.dot_product_attention
  • core.transformer.utils
• core.inference
  • core.inference.text_generation_controllers
    • core.inference.text_generation_controllers.vlm_text_generation_controller
    • core.inference.text_generation_controllers.encoder_decoder_text_generation_controller
    • core.inference.text_generation_controllers.simple_text_generation_controller
    • core.inference.text_generation_controllers.text_generation_controller
  • core.inference.model_inference_wrappers
    • core.inference.model_inference_wrappers.gpt
    • core.inference.model_inference_wrappers.t5
    • core.inference.model_inference_wrappers.inference_wrapper_config
    • core.inference.model_inference_wrappers.abstract_model_inference_wrapper
  • core.inference.contexts
    • core.inference.contexts.base_context
    • core.inference.contexts.dynamic_block_allocator
    • core.inference.contexts.static_context
    • core.inference.contexts.dynamic_context
    • core.inference.contexts.fused_kv_append_kernel
  • core.inference.text_generation_server
    • core.inference.text_generation_server.text_generation_server
    • core.inference.text_generation_server.tokenization
    • core.inference.text_generation_server.run_mcore_engine
  • core.inference.engines
    • core.inference.engines.abstract_engine
    • core.inference.engines.mcore_engine
    • core.inference.engines.async_zmq_communicator
    • core.inference.engines.dynamic_engine
    • core.inference.engines.static_engine
  • core.inference.batch_dimensions_utils
  • core.inference.inference_client
  • core.inference.communication_utils
  • core.inference.data_parallel_inference_coordinator
  • core.inference.unified_memory
  • core.inference.common_inference_params
  • core.inference.inference_request
  • core.inference.headers
  • core.inference.scheduler
  • core.inference.async_stream
  • core.inference.sampling_params
  • core.inference.utils
• core.optimizer
  • core.optimizer.cpu_offloading
    • core.optimizer.cpu_offloading.hybrid_optimizer
  • core.optimizer.clip_grads
  • core.optimizer.optimizer
  • core.optimizer.grad_scaler
  • core.optimizer.qk_clip
  • core.optimizer.optimizer_config
  • core.optimizer.distrib_optimizer
• core.pipeline_parallel
  • core.pipeline_parallel.bridge_communicator
  • core.pipeline_parallel.schedules
  • core.pipeline_parallel.p2p_communication
  • core.pipeline_parallel.combined_1f1b
  • core.pipeline_parallel.utils
• core.post_training
  • core.post_training.modelopt
    • core.post_training.modelopt.gpt
    • core.post_training.modelopt.mamba
    • core.post_training.modelopt.layers
• core.fusions
  • core.fusions.fused_weighted_squared_relu
  • core.fusions.fused_softmax
  • core.fusions.fused_pad_routing_map
  • core.fusions.fused_cross_entropy
  • core.fusions.fused_bias_dropout
  • core.fusions.fused_bias_geglu
  • core.fusions.fused_indices_converter
  • core.fusions.fused_layer_norm
  • core.fusions.fused_bias_swiglu
  • core.fusions.fused_bias_gelu
  • core.fusions.fused_mla_yarn_rope_apply

Submodules

• core.packed_seq_params
• core.rerun_state_machine
• core.config_logger
• core.energy_monitor
• core.activations
• core.safe_globals
• core.fp4_utils
• core.num_microbatches_calculator
• core.full_cuda_graph
• core.parallel_state
• core.inference_params
• core.process_groups_config
• core.jit
• core.msc_utils
• core.nccl_allocator
• core.timers
• core.package_info
• core.config
• core.model_parallel_config
• core.enums
• core.optimizer_param_scheduler
• core.hyper_comm_grid
• core.fp8_utils
• core.utils

Package Contents

Data

mpu
__all__

API

core.mpu

None

core.__all__

[‘parallel_state’, ‘tensor_parallel’, ‘utils’, ‘DistributedDataParallel’, ‘InferenceParams’, ‘ModelP…