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 15, 11:00 UTC
[Bugfix][V1] Fix FlashInfer V1 backend using the wrong VllmConfig (#18086)
Michael Goin committed
May 13, 2025 at 23:36 UTC
12e6c0b41c197c5079d6ef285930e0dd3bb60c37
1 file changed
+2
-3
vllm/v1/attention/backends/flashinfer.py
+2
-3
@@ -14,8 +14,7 @@ import vllm.envs as envs
14
from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl,
15
AttentionType)
16
from vllm.attention.layer import Attention
17
-from vllm.config import (VllmConfig, get_current_vllm_config,
18
- get_layers_from_vllm_config)
17
+from vllm.config import VllmConfig, get_layers_from_vllm_config
18
from vllm.logger import init_logger
19
from vllm.v1.attention.backends.flash_attn import use_cascade_attention
20
from vllm.v1.attention.backends.utils import CommonAttentionMetadata
@@ -215,7 +214,7 @@ class FlashInferMetadataBuilder:
214
# Global hyperparameters shared by all attention layers
215
self.global_hyperparameters: Optional[PerLayerParameters] = None
216
218
- self.vllm_config = get_current_vllm_config()
217
+ self.vllm_config = runner.vllm_config
218
self.kv_cache_spec = kv_cache_spec
219
self.block_table = block_table
220