Sharding

Module: generative_models.scaling.sharding

Source: generative_models/scaling/sharding.py

Overview

Sharding strategies and parallelism configuration for scalable training.

This module exposes the retained sharding strategies and configuration objects used by artifex.generative_models.scaling. It focuses on explicit strategy composition and mesh metadata instead of claiming a generic parameter-name to PartitionSpec inference layer.

Classes

DataParallelStrategy

class DataParallelStrategy

FSDPStrategy

class FSDPStrategy

MultiDimensionalStrategy

class MultiDimensionalStrategy

ParallelismConfig

class ParallelismConfig

PipelineParallelStrategy

class PipelineParallelStrategy

ShardingConfig

class ShardingConfig

ShardingStrategy

class ShardingStrategy

TensorParallelStrategy

class TensorParallelStrategy

Functions

init

def __init__()

init

def __init__()

init

def __init__()

init

def __init__()

init

def __init__()

apply_sharding

def apply_sharding()

apply_sharding

def apply_sharding()

apply_sharding

def apply_sharding()

apply_sharding

def apply_sharding()

apply_sharding

def apply_sharding()

assign_layers_to_stages

def assign_layers_to_stages()

from_device_count

def from_device_count()

from_sharding_config

def from_sharding_config()

get_attention_output_spec

def get_attention_output_spec()

get_attention_qkv_spec

def get_attention_qkv_spec()

get_backward_communication_pattern

def get_backward_communication_pattern()

get_combined_partition_spec

def get_combined_partition_spec()

get_forward_communication_pattern

def get_forward_communication_pattern()

get_gradient_partition_spec

def get_gradient_partition_spec()

get_linear_weight_spec

def get_linear_weight_spec()

get_partition_spec

def get_partition_spec()

get_partition_spec

def get_partition_spec()

get_partition_spec

def get_partition_spec()

get_partition_spec

def get_partition_spec()

get_partition_spec

def get_partition_spec()

get_sharding_constraints

def get_sharding_constraints()

get_total_device_count

def get_total_device_count()

is_valid

def is_valid()

resolve_sharding_conflicts

def resolve_sharding_conflicts()

should_shard_weight

def should_shard_weight()

Module Statistics

• Classes: 8
• Functions: 31
• Imports: 8