xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-30 08:57:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge 5ce77694018bd9b17d82318b8fe8b84f5453fe78 into 254f6b986720c92ddf97fbb1a6a6465da8e87e29

Browse Source
This commit is contained in:
Chauncey 2025-12-25 00:06:49 +00:00 committed by GitHub
commit 210add25a1
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 21 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/config/speculative.py
View File

@ -618,7 +618,7 @@ class SpeculativeConfig:
f"{self.disable_by_batch_size=}" f"{self.disable_by_batch_size=}"
) )
eagle3_target_supported = ["llama", "qwen", "minicpm", "gpt_oss"] eagle3_target_supported = ["llama", "qwen", "minicpm", "gpt_oss", "kimi_k2"]
if ( if (
self.method == "eagle3" self.method == "eagle3"
and self.target_model_config and self.target_model_config

22
vllm/model_executor/models/deepseek_v2.py
View File

@ -85,7 +85,13 @@ from vllm.v1.attention.backends.mla.indexer import (
from vllm.v1.kv_cache_interface import KVCacheSpec, MLAAttentionSpec from vllm.v1.kv_cache_interface import KVCacheSpec, MLAAttentionSpec
from vllm.v1.worker.workspace import current_workspace_manager from vllm.v1.worker.workspace import current_workspace_manager
from .interfaces import MixtureOfExperts, SupportsEagle, SupportsLoRA, SupportsPP from .interfaces import (
MixtureOfExperts,
SupportsEagle,
SupportsEagle3,
SupportsLoRA,
SupportsPP,
)
from .utils import ( from .utils import (
PPMissingLayer, PPMissingLayer,
is_pp_missing_parameter, is_pp_missing_parameter,
@ -1381,7 +1387,12 @@ class DeepseekV2MixtureOfExperts(MixtureOfExperts):
class DeepseekV2ForCausalLM( class DeepseekV2ForCausalLM(
nn.Module, SupportsPP, DeepseekV2MixtureOfExperts, SupportsLoRA, SupportsEagle nn.Module,
SupportsPP,
DeepseekV2MixtureOfExperts,
SupportsLoRA,
SupportsEagle,
SupportsEagle3,
): ):
packed_modules_mapping = { packed_modules_mapping = {
"gate_up_proj": ["gate_proj", "up_proj"], "gate_up_proj": ["gate_proj", "up_proj"],
@ -1463,6 +1474,13 @@ class DeepseekV2ForCausalLM(
def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor: def embed_input_ids(self, input_ids: torch.Tensor) -> torch.Tensor:
return self.model.embed_input_ids(input_ids) return self.model.embed_input_ids(input_ids)
def set_aux_hidden_state_layers(self, layers: tuple[int, ...]) -> None:
self.model.aux_hidden_state_layers = layers
def get_eagle3_aux_hidden_state_layers(self) -> tuple[int, ...]:
num_layers = len(self.model.layers)
return (2, num_layers // 2, num_layers - 3)
def forward( def forward(
self, self,
input_ids: torch.Tensor, input_ids: torch.Tensor,