biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V1]SupportsV0Only protocol for model definitions (#13959)

Browse Source 
Signed-off-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Roger Wang
2025-02-27 17:02:15 -08:00
committed by GitHub
parent 67fc426845
commit 6c85da3a18
19 changed files with 93 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/models/mamba2.py
View File

@@ -22,7 +22,8 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
DEFAULT_VOCAB_PADDING_SIZE, ParallelLMHead, VocabParallelEmbedding)
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.models.interfaces import (HasInnerState,
IsAttentionFree)
IsAttentionFree,
SupportsV0Only)
from vllm.model_executor.models.mamba_cache import (MambaCacheManager,
MambaCacheParams)
from vllm.model_executor.sampling_metadata import SamplingMetadata
@@ -174,7 +175,8 @@ class Mamba2Model(nn.Module):
return hidden_states
class Mamba2ForCausalLM(nn.Module, HasInnerState, IsAttentionFree):
class Mamba2ForCausalLM(nn.Module, HasInnerState, IsAttentionFree,
SupportsV0Only):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
config = vllm_config.model_config.hf_config