[Refactor] Remove deepseek dead code (#48780)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Wentao Ye
2026-07-17 14:57:13 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent c4dd6d78fd
commit 877dae9c68
7 changed files with 3 additions and 387 deletions
@@ -12,7 +12,6 @@ from transformers import DeepseekV2Config, DeepseekV3Config
from vllm.compilation.decorators import support_torch_compile
from vllm.config import VllmConfig, get_current_vllm_config
from vllm.logger import init_logger
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import ReplicatedLinear
from vllm.model_executor.layers.logits_processor import LogitsProcessor
@@ -36,8 +35,6 @@ from .utils import (
process_eagle_weight,
)
logger = init_logger(__name__)
class DeepseekV2Eagle3DecoderLayer(nn.Module):
"""
@@ -11,7 +11,6 @@ from vllm._aiter_ops import rocm_aiter_ops
from vllm.compilation.decorators import support_torch_compile
from vllm.config import VllmConfig
from vllm.distributed import tensor_model_parallel_all_gather
from vllm.logger import init_logger
from vllm.model_executor.layers.fused_moe import (
fused_moe_make_expert_params_mapping,
)
@@ -38,8 +37,6 @@ from .deepseek_v2 import (
)
from .utils import get_pp_missing_layer_names, maybe_prefix
logger = init_logger(__name__)
def _restore_full_token_layout_if_needed(
hidden_states: torch.Tensor,
@@ -15,7 +15,6 @@ import torch
from vllm.logger import init_logger
from vllm.tracing import instrument
from vllm.utils.math_utils import cdiv
logger = init_logger(__name__)
@@ -39,23 +38,6 @@ _DEFAULT_TOKEN_SIZE_CANDIDATES = (
)
def _compute_mhc_pre_num_split(
*,
num_tokens: int,
hidden_size: int,
hc_mult: int,
num_sms: int,
) -> int:
block_k = 64
block_m = 64
k = hc_mult * hidden_size
grid_size = cdiv(num_tokens, block_m)
split_k = num_sms // grid_size
num_block_k = cdiv(k, block_k)
split_k = min(split_k, num_block_k // 4)
return max(split_k, 1)
def _normalize_token_sizes(
token_sizes: Iterable[int],
*,