[Model] Add Kimi K3 support: model files and kernels [1/N] (#50089)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> Signed-off-by: Isotr0py <Isotr0py@outlook.com> Signed-off-by: Bugen Zhao <i@bugenzhao.com> Signed-off-by: Yongye Zhu <zyy1102000@gmail.com> Co-authored-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg> Co-authored-by: Bugen Zhao <i@bugenzhao.com> Co-authored-by: Yongye Zhu <zyy1102000@gmail.com> Co-authored-by: Ziming Huang <zelda.huanghuang@gmail.com> Co-authored-by: Roger Wang <hey@rogerw.io> Co-authored-by: Isotr0py <mozf@inferact.ai> Co-authored-by: aoshen02 <aoshen02@users.noreply.github.com> Co-authored-by: Woosuk Kwon <woosuk@inferact.ai> Co-authored-by: Jee Jee Li <jeejeelee@inferact.ai> Co-authored-by: aoshen02 <aoshen@inferact.ai> Co-authored-by: Summer Yang <girasoleyang@gmail.com> Co-authored-by: Kevin H. Luu <khluu000@gmail.com> Co-authored-by: Bowen Wang <abmfy@icloud.com> Co-authored-by: gnovack <novackgm@gmail.com> Co-authored-by: Nick Hill <nickhill123@gmail.com> Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai> Co-authored-by: xiaozhoupy <peiyuanzhou1994@gmail.com> Co-authored-by: Roy Wang <yasong.wang@inferact.ai> Co-authored-by: Jeff (Junze) Ma <93145857+majunze2001@users.noreply.github.com> Co-authored-by: Codex <codex@openai.com>
This commit is contained in:
co-authored by
Isotr0py
Thien Tran
Bugen Zhao
Yongye Zhu
Ziming Huang
Roger Wang
Isotr0py
aoshen02
Woosuk Kwon
Jee Jee Li
aoshen02
Summer Yang
Kevin H. Luu
Bowen Wang
gnovack
Nick Hill
Yifan Qiao
xiaozhoupy
Roy Wang
Jeff Ma
Codex
parent
6f00a1ae3b
commit
7c6729b769
@@ -0,0 +1,73 @@
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
||||
|
||||
import torch
|
||||
|
||||
from vllm.distributed import (
|
||||
get_tensor_model_parallel_rank,
|
||||
get_tensor_model_parallel_world_size,
|
||||
get_tp_group,
|
||||
tensor_model_parallel_all_gather,
|
||||
tensor_model_parallel_reduce_scatter,
|
||||
)
|
||||
|
||||
|
||||
def _custom_collective(
|
||||
name: str,
|
||||
x: torch.Tensor,
|
||||
) -> torch.Tensor | None:
|
||||
device_communicator = get_tp_group().device_communicator
|
||||
if device_communicator is None:
|
||||
return None
|
||||
collective = getattr(device_communicator, name, None)
|
||||
if collective is None:
|
||||
return None
|
||||
return collective(x)
|
||||
|
||||
|
||||
def sp_all_gather(x: torch.Tensor) -> torch.Tensor:
|
||||
output = _custom_collective("custom_all_gather", x)
|
||||
if output is not None:
|
||||
return output
|
||||
return tensor_model_parallel_all_gather(x, 0)
|
||||
|
||||
|
||||
def sp_reduce_scatter(x: torch.Tensor) -> torch.Tensor:
|
||||
assert x.ndim == 2
|
||||
tp_size = get_tensor_model_parallel_world_size()
|
||||
sp_pad = (-x.shape[0]) % tp_size
|
||||
if sp_pad > 0:
|
||||
x = torch.nn.functional.pad(x, (0, 0, 0, sp_pad))
|
||||
output = _custom_collective("custom_reduce_scatter", x)
|
||||
if output is not None:
|
||||
return output
|
||||
return tensor_model_parallel_reduce_scatter(x, 0)
|
||||
|
||||
|
||||
def sp_shard(x: torch.Tensor) -> torch.Tensor:
|
||||
assert x.ndim == 2
|
||||
tp_size = get_tensor_model_parallel_world_size()
|
||||
tp_rank = get_tensor_model_parallel_rank()
|
||||
sp_pad = (-x.shape[0]) % tp_size
|
||||
if sp_pad > 0:
|
||||
x = torch.nn.functional.pad(x, (0, 0, 0, sp_pad))
|
||||
chunk = x.shape[0] // tp_size
|
||||
return x[tp_rank * chunk : (tp_rank + 1) * chunk]
|
||||
|
||||
|
||||
def sp_padding_mask(
|
||||
is_padding: torch.Tensor | None,
|
||||
hidden_states: torch.Tensor,
|
||||
) -> torch.Tensor:
|
||||
num_tokens = hidden_states.shape[0]
|
||||
if is_padding is None:
|
||||
is_padding = hidden_states.new_zeros(num_tokens, dtype=torch.bool)
|
||||
assert is_padding.shape[0] == num_tokens
|
||||
|
||||
tp_size = get_tensor_model_parallel_world_size()
|
||||
sp_pad = (-num_tokens) % tp_size
|
||||
if sp_pad > 0:
|
||||
is_padding = torch.nn.functional.pad(is_padding, (0, sp_pad), value=True)
|
||||
chunk = is_padding.shape[0] // tp_size
|
||||
tp_rank = get_tensor_model_parallel_rank()
|
||||
return is_padding[tp_rank * chunk : (tp_rank + 1) * chunk]
|
||||
Reference in New Issue
Block a user