allreduce

Applicability

Product	Supported
Atlas A3 training products / Atlas A3 inference products	√
Atlas A2 training products / Atlas A2 inference products	√
Atlas 200I/500 A2 inference products	☓
Atlas inference products	√
Atlas training products	√

For the Atlas inference products , only the Atlas 300I Duo inference card is supported.

Description

Performs the reduction operation on the input data of all ranks in a group and sends the result to the output buffer of all ranks. The reduction operation type is specified by the reduction parameter. This API operates the collective communication operator AllReduce.

Prototype

      
           def allreduce(tensor, reduction, fusion=1, fusion_id=-1, group="hccl_world_group")

Parameters

Parameter	Input/Output	Description
tensor	Input	TensorFlow tensor type. For the Atlas A3 training products / Atlas A3 inference products , the supported data types are int8, int16, int32, int64, float16, float32, and bfp16. For the Atlas A2 training products / Atlas A2 inference products , the supported data types are int8, int16, int32, int64, float16, float32, and bfp16. Note that the performance will deteriorate for the int64 data type. For the Atlas training products , the supported data types are int8, int32, int64, float16, and float32. For the Atlas 300I Duo inference card, the supported data types are int8, int16, int32, float16, and float32.
reduction	Input	String type. Reduction operation types, which can include max, min, prod, and sum. NOTE: For the Atlas A3 training products / Atlas A3 inference products , the prod operation does not support the int16 and bfp16 data types in the current version. For the Atlas A2 training products / Atlas A2 inference products , the prod operation does not support the int16 and bfp16 data types in the current version. For the Atlas 300I Duo inference card, the prod, max, and min operations do not support the int16 data type in the current version.
fusion	Input	Int type. AllReduce operator fusion flag. The value can be one of the following: 0: The AllReduce operator is not fused with other AllReduce operators during network compilation. 1: The AllReduce operator is fused based on the gradient splitting policy during network compilation. 2: AllReduce operators with the same fusion_id are fused during network compilation.
fusion_id	Input	Int type. AllReduce operator fusion ID. When fusion is set to 2, AllReduce operators with the same fusion_id are fused during network compilation.
group	Input	A string containing a maximum of 128 bytes, including the end character. Group name, which can be a user-defined value or hccl_world_group.

Returns

The result tensor

Restrictions

The caller rank must be within the range defined by the group argument passed to this API call. Otherwise, the API call fails.
Each rank can have only one input.
The upstream node of allreduce must not be variable.
The input tensor size must be less than or equal to 8 GB.
For the AllReduce operator fusion, only the reduction type sum is supported.

Example

      
           from npu_bridge.hccl import hccl_ops
tensor = tf.random_uniform((1, 3), minval=1, maxval=10, dtype=tf.float32)
result = hccl_ops.allreduce(tensor, "sum")

Parent topic: npu_bridge.hccl.hccl_ops