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

[Hybrid] Added supports_mamba_prefix_caching Protocol (#27339)

Browse Source 
Signed-off-by: asafg <39553475+Josephasafg@users.noreply.github.com>
This commit is contained in:
Asaf Joseph Gardin
2025-10-27 15:05:20 +02:00
committed by GitHub
parent f4e8154076
commit 9273754222
10 changed files with 93 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/zamba2.py
View File

@@ -45,7 +45,7 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.sequence import IntermediateTensors
from .interfaces import HasInnerState, IsHybrid
from .interfaces import HasInnerState, IsHybrid, SupportsMambaPrefixCaching
from .utils import AutoWeightsLoader, WeightsMapper, maybe_prefix
@@ -824,7 +824,7 @@ class Zamba2Model(nn.Module):
return loaded_params
class Zamba2ForCausalLM(nn.Module, HasInnerState, IsHybrid):
class Zamba2ForCausalLM(nn.Module, HasInnerState, IsHybrid, SupportsMambaPrefixCaching):
"""Zamba2 model with causal language modeling head.
This class wraps the core Zamba2 model and adds: