Skip to content

[P/D] [Misc] Remove useless PD check in deepseek #2739

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Sep 4, 2025
Merged

[P/D] [Misc] Remove useless PD check in deepseek #2739

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
84336ef
PD disaggregate should not force with_prefill true in P and false in D
liziyu179 Sep 4, 2025
1b1da92
PD disaggregate should not force with_prefill false in D
liziyu179 Sep 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 1 addition & 12 deletions vllm_ascend/models/deepseek_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,8 +32,7 @@
from torch import nn
from transformers import PretrainedConfig
from vllm.attention import Attention, AttentionMetadata
from vllm.config import (CacheConfig, ModelConfig, VllmConfig,
get_current_vllm_config)
from vllm.config import CacheConfig, ModelConfig, VllmConfig
from vllm.distributed import (get_pp_group, get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size,
get_tp_group, split_tensor_along_last_dim,
Expand Down Expand Up @@ -374,10 +373,6 @@ def __init__(
self.tp_group = get_tp_group().device_group
self.tp_rank = get_tp_group().rank_in_group
self.ep_group = get_ep_group()
self.kv_consumer = None
transfer_config = get_current_vllm_config().kv_transfer_config
if transfer_config is not None:
self.kv_consumer = transfer_config.kv_role == "kv_consumer"

self.params_dtype = torch.get_default_dtype()
self.rm_router_logits = self.experts.rm_router_logits
Expand All @@ -395,12 +390,6 @@ def forward(self,

is_prefill = forward_context.with_prefill

# If this node is kv_consumer, we force the moe always runs in decode path to make sure
# the behaviour aligned between dummy_run and normal model_execute.
if self.kv_consumer:
is_prefill = False
enable_force_load_balance = False

# router_logits: (num_tokens, n_experts)
router_logits = None
if not self.rm_router_logits and not self.enable_multistream_moe:
Expand Down
Loading