A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 19:30 UTC
[Bugfix] Fix PaliGemma MMP (#6930)
Roger Wang committed
Jul 30, 2024 at 02:20 UTC
c66c7f86aca956014d9ec6cc7a3e6001037e4655
1 file changed
+2
-5
vllm/model_executor/models/paligemma.py
+2
-5
@@ -9,7 +9,6 @@ from vllm.attention import AttentionMetadata
9
from vllm.config import CacheConfig, MultiModalConfig
10
from vllm.inputs import INPUT_REGISTRY, InputContext, LLMInputs
11
from vllm.logger import init_logger
12
-from vllm.model_executor.layers.linear import ColumnParallelLinear
12
from vllm.model_executor.layers.logits_processor import LogitsProcessor
13
from vllm.model_executor.layers.quantization.base_config import (
14
QuantizationConfig)
@@ -133,12 +132,10 @@ class PaliGemmaMultiModalProjector(nn.Module):
132
def __init__(self, vision_hidden_size: int, projection_dim: int):
133
super().__init__()
134
136
- self.linear = ColumnParallelLinear(vision_hidden_size,
137
- projection_dim,
138
- bias=True)
135
+ self.linear = nn.Linear(vision_hidden_size, projection_dim, bias=True)
136
137
def forward(self, image_features: torch.Tensor) -> torch.Tensor:
141
- hidden_states, _ = self.linear(image_features)
138
+ hidden_states = self.linear(image_features)
139
return hidden_states
140
141