id int64 0 190k | prompt stringlengths 21 13.4M | docstring stringlengths 1 12k ⌀ |
|---|---|---|
8,962 | import itertools
from typing import List, Tuple
import torch
import torch._prims_common as utils
The provided code snippet includes necessary dependencies for implementing the `fill_defaults` function. Write a Python function `def fill_defaults(args, n, defaults_tail)` to solve the following problem:
__torch_dispatch_... | __torch_dispatch__ doesn't guarantee the number of arguments you are passed (e.g., defaulted arguments are not passed); but usually it is convenient to pad out the arguments list with defaults. This function helps you do that. Args: args: the list of positional arguments passed to __torch_dispatch__ n: the number of ar... |
8,963 | import itertools
from typing import List, Tuple
import torch
import torch._prims_common as utils
def find_arg_of_type(it, t):
for x in it:
if isinstance(x, t):
return x
return None | null |
8,964 | import copy
from collections import OrderedDict
from dataclasses import dataclass, field
from typing import (
Any,
cast,
Dict,
Iterator,
List,
Mapping,
Optional,
Set,
Tuple,
Type,
Union,
)
import torch
from torch import nn, Tensor
from torch.nn.modules.module import _Incompat... | null |
8,965 | import copy
from collections import OrderedDict
from dataclasses import dataclass, field
from typing import (
Any,
cast,
Dict,
Iterator,
List,
Mapping,
Optional,
Set,
Tuple,
Type,
Union,
)
import torch
from torch import nn, Tensor
from torch.nn.modules.module import _Incompat... | null |
8,966 | import copy
from collections import OrderedDict
from dataclasses import dataclass, field
from typing import (
Any,
cast,
Dict,
Iterator,
List,
Mapping,
Optional,
Set,
Tuple,
Type,
Union,
)
import torch
from torch import nn, Tensor
from torch.nn.modules.module import _Incompat... | null |
8,967 | import copy
from collections import OrderedDict
from dataclasses import dataclass, field
from typing import (
Any,
cast,
Dict,
Iterator,
List,
Mapping,
Optional,
Set,
Tuple,
Type,
Union,
)
import torch
from torch import nn, Tensor
from torch.nn.modules.module import _Incompat... | null |
8,968 | import copy
from dataclasses import dataclass
from typing import Any, Dict, List, Mapping, Optional, TypeVar, Union
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch.distributed import _remote_device
from torch.distributed._shard... | null |
8,969 | import copy
from dataclasses import dataclass
from typing import Any, Dict, List, Mapping, Optional, TypeVar, Union
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch.distributed import _remote_device
from torch.distributed._shard... | null |
8,970 | import copy
import logging
from collections import OrderedDict
from itertools import accumulate
from typing import Any, Dict, List, Optional, Set, Type, TypeVar, Union
import torch
from fbgemm_gpu.split_embedding_configs import EmbOptimType
from torch import nn
from torchrec import optim as trec_optim
from torchrec.dis... | Appends provided prefix to provided name. |
8,971 | import copy
import logging
from collections import OrderedDict
from itertools import accumulate
from typing import Any, Dict, List, Optional, Set, Type, TypeVar, Union
import torch
from fbgemm_gpu.split_embedding_configs import EmbOptimType
from torch import nn
from torchrec import optim as trec_optim
from torchrec.dis... | Filters state dict for keys that start with provided name. Strips provided name from beginning of key in the resulting state dict. Args: state_dict (OrderedDict[str, torch.Tensor]): input state dict to filter. name (str): name to filter from state dict keys. Returns: OrderedDict[str, torch.Tensor]: filtered state dict. |
8,972 | import copy
import logging
from collections import OrderedDict
from itertools import accumulate
from typing import Any, Dict, List, Optional, Set, Type, TypeVar, Union
import torch
from fbgemm_gpu.split_embedding_configs import EmbOptimType
from torch import nn
from torchrec import optim as trec_optim
from torchrec.dis... | Adds prefix to all keys in state dict, in place. Args: state_dict (Dict[str, Any]): input state dict to update. prefix (str): name to filter from state dict keys. Returns: None. |
8,973 | import copy
import logging
from collections import OrderedDict
from itertools import accumulate
from typing import Any, Dict, List, Optional, Set, Type, TypeVar, Union
import torch
from fbgemm_gpu.split_embedding_configs import EmbOptimType
from torch import nn
from torchrec import optim as trec_optim
from torchrec.dis... | Retrieves names of top level modules that do not contain any sharded sub-modules. Args: model (torch.nn.Module): model to retrieve unsharded module names from. Returns: List[str]: list of names of modules that don't have sharded sub-modules. |
8,974 | import copy
import logging
from collections import OrderedDict
from itertools import accumulate
from typing import Any, Dict, List, Optional, Set, Type, TypeVar, Union
import torch
from fbgemm_gpu.split_embedding_configs import EmbOptimType
from torch import nn
from torchrec import optim as trec_optim
from torchrec.dis... | null |
8,975 | import itertools
import logging
from typing import Callable, Dict, List, Optional
import torch
import torch.distributed as dist
from torch import nn
from torch.autograd.profiler import record_function
from torchrec.distributed.comm_ops import (
all_gather_base_pooled,
alltoall_pooled,
alltoall_sequence,
... | null |
8,976 | import itertools
import logging
from typing import Callable, Dict, List, Optional
import torch
import torch.distributed as dist
from torch import nn
from torch.autograd.profiler import record_function
from torchrec.distributed.comm_ops import (
all_gather_base_pooled,
alltoall_pooled,
alltoall_sequence,
... | Calculates relevant recat indices required to reorder AlltoAll collective. Args: local_split (int): number of features in local split. num_splits (int): number of splits (typically WORLD_SIZE). stagger (int): secondary reordering, (typically 1, but `WORLD_SIZE/LOCAL_WORLD_SIZE` for TWRW). device (Optional[torch.device]... |
8,977 | import abc
import operator
from dataclasses import dataclass
from enum import Enum, unique
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Type,
TypeVar,
Union,
)
from fbgemm_gpu.split_table_batched_embeddings_ops_common import (
BoundsC... | Format a table as a string. Parameters: table (list of lists or list of tuples): The data to be formatted as a table. headers (list of strings, optional): The column headers for the table. If not provided, the first row of the table will be used as the headers. Returns: str: A string representation of the table. |
8,978 | import abc
import operator
from dataclasses import dataclass
from enum import Enum, unique
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Type,
TypeVar,
Union,
)
from fbgemm_gpu.split_table_batched_embeddings_ops_common import (
BoundsC... | null |
8,979 | import abc
import operator
from dataclasses import dataclass
from enum import Enum, unique
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Type,
TypeVar,
Union,
)
from fbgemm_gpu.split_table_batched_embeddings_ops_common import (
BoundsC... | null |
8,980 | import copy
from collections import defaultdict, OrderedDict
from dataclasses import dataclass
from typing import Any, DefaultDict, Dict, Iterator, List, Optional, Type
import torch
import torch.distributed as dist
from torch import nn
from torch.distributed._shard.sharded_tensor import Shard
from torchrec.distributed.... | null |
8,981 | from collections import defaultdict, deque
from dataclasses import dataclass
from typing import Any, cast, Dict, List, Optional, Tuple, Type
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch import nn
from torchrec.distributed.em... | null |
8,982 | from collections import defaultdict, deque
from dataclasses import dataclass
from typing import Any, cast, Dict, List, Optional, Tuple, Type
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch import nn
from torchrec.distributed.em... | null |
8,983 | from collections import defaultdict, deque
from dataclasses import dataclass
from typing import Any, cast, Dict, List, Optional, Tuple, Type
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch import nn
from torchrec.distributed.em... | null |
8,984 | from typing import Any, Dict, Iterable, Optional
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torchrec.distributed.embedding_types import GroupedEmbeddingConfig
class TBEToRegisterMixIn:
def get_tbes_to_register(
self,... | null |
8,985 | from typing import Any, Dict, Iterable, Optional
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torchrec.distributed.embedding_types import GroupedEmbeddingConfig
FUSED_PARAM_REGISTER_TBE_BOOL: str = "__register_tbes_in_named_module... | null |
8,986 | from typing import Any, Dict, Iterable, Optional
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torchrec.distributed.embedding_types import GroupedEmbeddingConfig
FUSED_PARAM_QUANT_STATE_DICT_SPLIT_SCALE_BIAS: str = (
"__registe... | null |
8,987 | from typing import Any, Dict, Iterable, Optional
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torchrec.distributed.embedding_types import GroupedEmbeddingConfig
FUSED_PARAM_REGISTER_TBE_BOOL: str = "__register_tbes_in_named_module... | null |
8,988 | import logging
import os
from typing import List, Optional, Tuple
import torch
import torch.distributed as dist
logger: logging.Logger = logging.getLogger(__name__)
_INTRA_PG: Optional[dist.ProcessGroup] = None
_CROSS_PG: Optional[dist.ProcessGroup] = None
def get_local_size(world_size: Optional[int] = None) -> int:
... | Creates sub process groups (intra and cross node) |
8,989 | from functools import partial
from typing import Any, Dict, Iterator, List, Optional, Type, Union
import torch
from torch import nn
from torchrec.distributed.embedding_types import (
BaseEmbeddingSharder,
KJTList,
ShardedEmbeddingModule,
)
from torchrec.distributed.embeddingbag import (
EmbeddingBagColl... | null |
8,990 | import copy
import logging
import warnings
from collections import defaultdict, deque, OrderedDict
from dataclasses import dataclass, field
from itertools import accumulate
from typing import Any, cast, Dict, List, MutableMapping, Optional, Type, Union
import torch
from torch import nn
from torch.autograd.profiler impo... | null |
8,991 | import copy
import logging
import warnings
from collections import defaultdict, deque, OrderedDict
from dataclasses import dataclass, field
from itertools import accumulate
from typing import Any, cast, Dict, List, MutableMapping, Optional, Type, Union
import torch
from torch import nn
from torch.autograd.profiler impo... | null |
8,992 | import copy
import logging
import warnings
from collections import defaultdict, deque, OrderedDict
from dataclasses import dataclass, field
from itertools import accumulate
from typing import Any, cast, Dict, List, MutableMapping, Optional, Type, Union
import torch
from torch import nn
from torch.autograd.profiler impo... | null |
8,993 | import copy
import logging
import warnings
from collections import defaultdict, deque, OrderedDict
from dataclasses import dataclass, field
from itertools import accumulate
from typing import Any, cast, Dict, List, MutableMapping, Optional, Type, Union
import torch
from torch import nn
from torch.autograd.profiler impo... | null |
8,994 | import copy
from typing import Any, Dict, List, Optional, Type
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch import nn
from torchrec.distributed.embedding_lookup import EmbeddingComputeKernel
from torchrec.distributed.embeddi... | null |
8,995 | import copy
from typing import Any, Dict, List, Optional, Type
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch import nn
from torchrec.distributed.embedding_lookup import EmbeddingComputeKernel
from torchrec.distributed.embeddi... | null |
8,996 | import copy
from typing import Any, Dict, List, Optional, Type
import torch
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from torch import nn
from torchrec.distributed.embedding_lookup import EmbeddingComputeKernel
from torchrec.distributed.embeddi... | null |
8,997 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
8,998 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
8,999 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Performs AlltoAll operation for a single pooled embedding tensor. Each process splits the input pooled embeddings tensor based on the world size, and then scatters the split list to all processes in the group. Then concatenates the received tensors from all processes in the group and returns a single output tensor. Arg... |
9,000 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,001 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Performs AlltoAll operation for sequence embeddings. Each process splits the input tensor based on the world size, and then scatters the split list to all processes in the group. Then concatenates the received tensors from all processes in the group and returns a single output tensor. NOTE: AlltoAll operator for Sequen... |
9,002 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Performs `alltoallv` operation for a list of input embeddings. Each process scatters the list to all processes in the group. Args: inputs (List[Tensor]): list of tensors to scatter, one per rank. The tensors in the list usually have different lengths. out_split (Optional[List[int]]): output split sizes (or dim_sum_per_... |
9,003 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Performs reduce-scatter operation for a pooled embeddings tensor split into world size number of chunks. The result of the reduce operation gets scattered to all processes in the group. Args: inputs (List[Tensor]): list of tensors to scatter, one per rank. group (Optional[dist.ProcessGroup]): the process group to work ... |
9,004 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Reduces then scatters a flattened pooled embeddings tensor to all processes in a group. Input tensor is of size `output_tensor_size * world_size`. Args: input (Tensor): flattened tensor to scatter. group (Optional[dist.ProcessGroup]): the process group to work on. If None, the default process group will be used. codecs... |
9,005 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | All-gathers tensors from all processes in a group to form a flattened pooled embeddings tensor. Input tensor is of size `output_tensor_size / world_size`. Args: input (Tensor): tensor to gather. group (Optional[dist.ProcessGroup]): the process group to work on. If None, the default process group will be used. Returns: ... |
9,006 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Performs reduce-scatter-v operation for a pooled embeddings tensor split unevenly into world size number of chunks. The result of the reduce operation gets scattered to all processes in the group according to `input_splits`. Args: input (Tensor): tensor to scatter. input_splits (List[int]): input splits. group (Optiona... |
9,007 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | Performs reduce-scatter-v operation for a 1-d pooled embeddings tensor of variable batch size per feature split unevenly into world size number of chunks. The result of the reduce operation gets scattered to all processes in the group. Args: input (Tensor): tensors to scatter, one per rank. batch_size_per_rank_per_feat... |
9,008 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,009 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,010 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,011 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,012 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,013 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,014 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,015 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,016 | from dataclasses import dataclass, field
from typing import Any, List, Optional, Tuple, TypeVar
import torch
import torch.distributed as dist
from torch import Tensor
from torch.autograd import Function
from torch.autograd.profiler import record_function
from torchrec.distributed.types import Awaitable, NoWait, Quantiz... | null |
9,017 | import abc
import copy
import itertools
from dataclasses import dataclass
from typing import (
Any,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Tuple,
TypeVar,
Union,
)
import torch
import torch.distributed as dist
from fbgemm_gpu.split_table_batched_embeddings_ops_inference i... | null |
9,018 | import abc
import copy
import itertools
from dataclasses import dataclass
from typing import (
Any,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Tuple,
TypeVar,
Union,
)
import torch
import torch.distributed as dist
from fbgemm_gpu.split_table_batched_embeddings_ops_inference i... | null |
9,019 | from typing import Callable, Dict, List, Optional, Type, Union
import torch
import torch.distributed as dist
from torch import nn
from torch.distributed._composable.contract import contract
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.model_parallel import get_default_sharders
from tor... | Replaces all sub_modules that are embedding modules with their sharded variants. This embedding_module -> sharded_embedding_module mapping is derived from the passed in sharders. This will leave the other parts of the model unaffected. It returns the original module Args: module (nn.Module): module to wrap. env (Option... |
9,020 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Returns a generator of ParameterShardingPlan for `ShardingType::DATA_PARALLEL` for construct_module_sharding_plan. Example:: ebc = EmbeddingBagCollection(...) plan = construct_module_sharding_plan( ebc, { "table_0": data_parallel(), }, ) |
9,021 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Returns a generator of ParameterShardingPlan for `ShardingType::TABLE_WISE` for construct_module_sharding_plan. Args: rank (int): rank to place table when doing table wise Example:: ebc = EmbeddingBagCollection(...) plan = construct_module_sharding_plan( ebc, { "table_0": table_wise(rank=0), }, ) |
9,022 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Returns a generator of ParameterShardingPlan for `ShardingType::ROW_WISE` for construct_module_sharding_plan. Args: sizes_placement (Optional[Tuple[List[int], str]]): Only use it in inference for uneven shardinglist of tuples of (sizes, placement); sizes is the row size list Example:: ebc = EmbeddingBagCollection(...) ... |
9,023 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Returns a generator of ParameterShardingPlan for `ShardingType::COLUMN_WISE` for construct_module_sharding_plan. Table will the sharded column-wise evenly across specified ranks (and can reuse ranks). Args: ranks (List[int]): ranks to place columns Example:: ebc = EmbeddingBagCollection(...) plan = construct_module_sha... |
9,024 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Returns a generator of ParameterShardingPlan for `ShardingType::TABLE_ROW_WISE` for construct_module_sharding_plan. Args: host_index (int): index of host (node) to do row wise Example:: ebc = EmbeddingBagCollection(...) plan = construct_module_sharding_plan( ebc, { "table_4": table_row_wise(host_index=2), }, ) |
9,025 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Convenience function to apply a sharding scheme generator for all modules in construct_module_sharding_plan. Example:: ebc = EmbeddingBagCollection(...) sharder = EmbeddingBagCollectionSharder() plan = construct_parameter_sharding_plan( ebc, apply_to_all(ebc, row_wise(), sharder), ) |
9,026 | import math
import warnings
from typing import Callable, cast, Dict, List, Optional, Tuple, Type
import torch
from torch import distributed as dist, nn
from torchrec.distributed.comm import get_local_size
from torchrec.distributed.embedding import EmbeddingCollectionSharder
from torchrec.distributed.embedding_types imp... | Helper function to create module sharding plans (EmbeddingModuleShardingPlan) for an module Args: module (nn.Module): module to create plan for. per_param_sharding: Dict[str, Callable[[nn.Parameter, int, int, str], ParameterSharding]]: A mapping of parameter names to a generator function that takes in [parameter, local... |
9,027 | import abc
import copy
from collections import OrderedDict
from typing import Any, cast, Dict, Iterator, List, Optional, Set, Tuple, Type
import torch
import torch.distributed as dist
from torch import nn
from torch.distributed.algorithms.ddp_comm_hooks import (
default_hooks as ddp_default_hooks,
)
from torch.dist... | Unwraps module wrapped by DMP, DDP, or FSDP. |
9,028 | import abc
import copy
from collections import OrderedDict
from typing import Any, cast, Dict, Iterator, List, Optional, Set, Tuple, Type
import torch
import torch.distributed as dist
from torch import nn
from torch.distributed.algorithms.ddp_comm_hooks import (
default_hooks as ddp_default_hooks,
)
from torch.dist... | Unwraps DMP module. Does not unwrap data parallel wrappers (i.e. DDP/FSDP), so overriding implementations by the wrappers can be used. |
9,029 | from typing import List, Tuple
from torchrec.distributed.quant_embeddingbag import ShardedQuantEmbeddingBagCollection
class ShardedQuantEmbeddingBagCollection(
ShardedQuantEmbeddingModuleState[
ListOfKJTList,
List[List[torch.Tensor]],
KeyedTensor,
NullShardedModuleContext,
],
):... | null |
9,030 | import abc
import copy
import uuid
from collections import defaultdict
from dataclasses import dataclass
from typing import Any, Dict, Generic, List, Optional, Tuple, TypeVar, Union
import torch
from torch import distributed as dist, nn
from torchrec.distributed.dist_data import (
KJTAllToAllTensorsAwaitable,
S... | Bucketizes the `values` in KeyedJaggedTensor into `num_buckets` buckets, `lengths` are readjusted based on the bucketization results. Note: This function should be used only for row-wise sharding before calling `KJTAllToAll`. Args: num_buckets (int): number of buckets to bucketize the values into. block_sizes: (torch.T... |
9,031 | import abc
import copy
import uuid
from collections import defaultdict
from dataclasses import dataclass
from typing import Any, Dict, Generic, List, Optional, Tuple, TypeVar, Union
import torch
from torch import distributed as dist, nn
from torchrec.distributed.dist_data import (
KJTAllToAllTensorsAwaitable,
S... | Groups tables by `DataType`, `PoolingType`, and `EmbeddingComputeKernel`. Args: tables_per_rank (List[List[ShardedEmbeddingTable]]): list of sharded embedding tables per rank with consistent weightedness. Returns: List[List[GroupedEmbeddingConfig]]: per rank list of GroupedEmbeddingConfig for features. |
9,032 | import abc
import copy
import uuid
from collections import defaultdict
from dataclasses import dataclass
from typing import Any, Dict, Generic, List, Optional, Tuple, TypeVar, Union
import torch
from torch import distributed as dist, nn
from torchrec.distributed.dist_data import (
KJTAllToAllTensorsAwaitable,
S... | null |
9,033 | import abc
import copy
import uuid
from collections import defaultdict
from dataclasses import dataclass
from typing import Any, Dict, Generic, List, Optional, Tuple, TypeVar, Union
import torch
from torch import distributed as dist, nn
from torchrec.distributed.dist_data import (
KJTAllToAllTensorsAwaitable,
S... | null |
9,034 | import abc
import copy
import uuid
from collections import defaultdict
from dataclasses import dataclass
from typing import Any, Dict, Generic, List, Optional, Tuple, TypeVar, Union
import torch
from torch import distributed as dist, nn
from torchrec.distributed.dist_data import (
KJTAllToAllTensorsAwaitable,
S... | null |
9,035 | import logging
from abc import ABC
from collections import OrderedDict
from typing import Any, cast, Dict, Iterator, List, Optional, Tuple, Union
import torch
import torch.distributed as dist
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from fbgemm... | null |
9,036 | import logging
from abc import ABC
from collections import OrderedDict
from typing import Any, cast, Dict, Iterator, List, Optional, Tuple, Union
import torch
import torch.distributed as dist
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from fbgemm... | null |
9,037 | import logging
from abc import ABC
from collections import OrderedDict
from typing import Any, cast, Dict, Iterator, List, Optional, Tuple, Union
import torch
import torch.distributed as dist
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from fbgemm... | null |
9,038 | import logging
from abc import ABC
from collections import OrderedDict
from typing import Any, cast, Dict, Iterator, List, Optional, Tuple, Union
import torch
import torch.distributed as dist
from fbgemm_gpu.split_table_batched_embeddings_ops_inference import (
IntNBitTableBatchedEmbeddingBagsCodegen,
)
from fbgemm... | null |
9,039 | import copy
import itertools
import logging
from collections import defaultdict
from dataclasses import dataclass, field
from threading import Event, Thread
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Set,
Tuple,
Type,
TypeVar,
U... | null |
9,040 | import copy
import itertools
import logging
from collections import defaultdict
from dataclasses import dataclass, field
from threading import Event, Thread
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Set,
Tuple,
Type,
TypeVar,
U... | As mentioned in https://pytorch.org/docs/stable/generated/torch.Tensor.record_stream.html, PyTorch uses the "caching allocator" for memory allocation for tensors. When a tensor is freed, its memory is likely to be reused by newly constructed tenosrs. By default, this allocator traces whether a tensor is still in use by... |
9,041 | import copy
import itertools
import logging
from collections import defaultdict
from dataclasses import dataclass, field
from threading import Event, Thread
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Set,
Tuple,
Type,
TypeVar,
U... | null |
9,042 | import copy
import itertools
import logging
from collections import defaultdict
from dataclasses import dataclass, field
from threading import Event, Thread
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Set,
Tuple,
Type,
TypeVar,
U... | null |
9,043 | import copy
import itertools
import logging
from collections import defaultdict
from dataclasses import dataclass, field
from threading import Event, Thread
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Set,
Tuple,
Type,
TypeVar,
U... | Overrides each input dist forward to support fusing the splits collective. NOTE: this can only be called after the input dists are initialized. |
9,044 | import copy
import itertools
import logging
from collections import defaultdict
from dataclasses import dataclass, field
from threading import Event, Thread
from typing import (
Any,
Callable,
cast,
Dict,
Generic,
Iterator,
List,
Optional,
Set,
Tuple,
Type,
TypeVar,
U... | null |
9,045 | import abc
import logging
from typing import cast, Generic, Iterator, List, Optional, Tuple
import torch
from torch.autograd.profiler import record_function
from torchrec.distributed.model_parallel import ShardedModule
from torchrec.distributed.train_pipeline.utils import (
_override_input_dist_forwards,
_rewri... | null |
9,046 | from functools import wraps
from typing import Any, Callable, cast, Optional, TypeVar
import torch.distributed as dist
The provided code snippet includes necessary dependencies for implementing the `is_leader` function. Write a Python function `def is_leader(pg: Optional[dist.ProcessGroup], leader_rank: int = 0) -> bo... | Checks if the current processs is the leader. Args: pg (Optional[dist.ProcessGroup]): the process's rank within the pg is used to determine if the process is the leader. pg being None implies that the process is the only member in the group (e.g. a single process program). leader_rank (int): the definition of leader (d... |
9,047 | from functools import wraps
from typing import Any, Callable, cast, Optional, TypeVar
import torch.distributed as dist
T = TypeVar("T")
def invoke_on_rank_and_broadcast_result(
pg: dist.ProcessGroup,
rank: int,
func: Callable[..., T],
*args: Any,
**kwargs: Any,
) -> T:
"""
Invokes a function... | null |
9,048 | import copy
import itertools
import logging
from decimal import Decimal
from typing import cast, Dict, List, Optional, Set, Tuple, Union
import torch
from torchrec.distributed.embedding_types import EmbeddingComputeKernel
from torchrec.distributed.planner.types import (
Enumerator,
Perf,
Proposer,
Shard... | null |
9,049 | import copy
import itertools
import logging
from decimal import Decimal
from typing import cast, Dict, List, Optional, Set, Tuple, Union
import torch
from torchrec.distributed.embedding_types import EmbeddingComputeKernel
from torchrec.distributed.planner.types import (
Enumerator,
Perf,
Proposer,
Shard... | only works for static_feedback proposers (the path of proposals to check is independent of the performance of the proposals) |
9,050 | import copy
import logging
import math
from typing import Dict, List, Optional, Set, Tuple
from torch import nn
from torchrec.distributed.planner.constants import BIGINT_DTYPE, POOLING_FACTOR
from torchrec.distributed.planner.types import (
ParameterConstraints,
PlannerError,
PlannerErrorType,
Storage,
... | null |
9,051 | import copy
import logging
import math
from typing import Dict, List, Optional, Set, Tuple
from torch import nn
from torchrec.distributed.planner.constants import BIGINT_DTYPE, POOLING_FACTOR
from torchrec.distributed.planner.types import (
ParameterConstraints,
PlannerError,
PlannerErrorType,
Storage,
... | null |
9,052 | import copy
import logging
import math
from typing import Dict, List, Optional, Set, Tuple
from torch import nn
from torchrec.distributed.planner.constants import BIGINT_DTYPE, POOLING_FACTOR
from torchrec.distributed.planner.types import (
ParameterConstraints,
PlannerError,
PlannerErrorType,
Storage,
... | null |
9,053 | import copy
import logging
import math
from typing import Dict, List, Optional, Set, Tuple
from torch import nn
from torchrec.distributed.planner.constants import BIGINT_DTYPE, POOLING_FACTOR
from torchrec.distributed.planner.types import (
ParameterConstraints,
PlannerError,
PlannerErrorType,
Storage,
... | null |
9,054 | import math
import operator
from functools import reduce
from typing import Any, cast, Dict, Iterable, List, Optional, Tuple, Type, Union
import torch
from torchrec.distributed.planner.types import Perf, ShardingOption, Storage
from torchrec.distributed.types import ShardingType
def bytes_to_mb(num_bytes: Union[float,... | null |
9,055 | import math
import operator
from functools import reduce
from typing import Any, cast, Dict, Iterable, List, Optional, Tuple, Type, Union
import torch
from torchrec.distributed.planner.types import Perf, ShardingOption, Storage
from torchrec.distributed.types import ShardingType
def gb_to_bytes(gb: float) -> int:
... | null |
9,056 | import math
import operator
from functools import reduce
from typing import Any, cast, Dict, Iterable, List, Optional, Tuple, Type, Union
import torch
from torchrec.distributed.planner.types import Perf, ShardingOption, Storage
from torchrec.distributed.types import ShardingType
def bytes_to_gb(num_bytes: int) -> float... | null |
9,057 | import math
import operator
from functools import reduce
from typing import Any, cast, Dict, Iterable, List, Optional, Tuple, Type, Union
import torch
from torchrec.distributed.planner.types import Perf, ShardingOption, Storage
from torchrec.distributed.types import ShardingType
class ShardingOption:
"""
One w... | null |
9,058 | import math
import operator
from functools import reduce
from typing import Any, cast, Dict, Iterable, List, Optional, Tuple, Type, Union
import torch
from torchrec.distributed.planner.types import Perf, ShardingOption, Storage
from torchrec.distributed.types import ShardingType
class Perf:
"""
Representation ... | Find the tables that are causing the imbalance, and return their names. |
9,059 | import copy
from functools import reduce
from time import perf_counter
from typing import cast, Dict, List, Optional, Tuple, Union
import torch
import torch.distributed as dist
from torch import nn
from torchrec.distributed.collective_utils import invoke_on_rank_and_broadcast_result
from torchrec.distributed.comm impor... | null |
9,060 | import logging
from collections import defaultdict
from typing import Any, cast, Dict, List, Optional, Tuple, Union
from torch import nn
from torchrec.distributed.planner.constants import BIGINT_DTYPE, NUM_POOLINGS
from torchrec.distributed.planner.shard_estimators import _calculate_shard_io_sizes
from torchrec.distrib... | null |
9,061 | import logging
from collections import defaultdict
from typing import Any, cast, Dict, List, Optional, Tuple, Union
from torch import nn
from torchrec.distributed.planner.constants import BIGINT_DTYPE, NUM_POOLINGS
from torchrec.distributed.planner.shard_estimators import _calculate_shard_io_sizes
from torchrec.distrib... | null |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.