[gpt-oss] Add gpt-oss bf16 support
This commit is contained in:
20
vllm/distributed/device_communicators/neuron_communicator.py
Normal file
20
vllm/distributed/device_communicators/neuron_communicator.py
Normal file
@@ -0,0 +1,20 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
import torch
|
||||
|
||||
from vllm.distributed.device_communicators.base_device_communicator import (
|
||||
DeviceCommunicatorBase)
|
||||
from vllm.platforms import current_platform
|
||||
|
||||
if current_platform.is_neuron():
|
||||
import torch_xla.core.xla_model as xm
|
||||
|
||||
|
||||
class NeuronCommunicator(DeviceCommunicatorBase):
|
||||
|
||||
def all_reduce(self, x: torch.Tensor) -> torch.Tensor:
|
||||
return xm.all_reduce(xm.REDUCE_SUM, x)
|
||||
|
||||
def all_gather(self, x: torch.Tensor, dim: int = -1) -> torch.Tensor:
|
||||
assert dim == -1, "Neuron only supports dim=-1 for all-gather."
|
||||
return xm.all_gather(x, dim=dim)
|
||||
Reference in New Issue
Block a user