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, 21:00 UTC
[V1][Minor] Do not print attn backend twice (#13985)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Woosuk Kwon committed
Feb 28, 2025 at 23:09 UTC
3b5567a2099ee8c6a153ad5cf61397b99d200eb6
1 file changed
+4
-3
vllm/platforms/cuda.py
+4
-3
@@ -178,7 +178,8 @@ class CudaPlatformBase(Platform):
178
block_size)
179
else:
180
if use_v1:
181
- logger.info("Using FlashMLA backend on V1 engine.")
181
+ logger.info_once(
182
+ "Using FlashMLA backend on V1 engine.")
183
return ("vllm.v1.attention.backends.mla."
184
"flashmla.FlashMLABackend")
185
else:
@@ -187,14 +188,14 @@ class CudaPlatformBase(Platform):
188
"flashmla.FlashMLABackend")
189
190
if use_v1:
190
- logger.info("Using Triton MLA backend on V1 engine.")
191
+ logger.info_once("Using Triton MLA backend on V1 engine.")
192
return ("vllm.v1.attention.backends.mla."
193
"triton_mla.TritonMLABackend")
194
else:
195
logger.info("Using Triton MLA backend.")
196
return "vllm.attention.backends.triton_mla.TritonMLABackend"
197
if use_v1:
197
- logger.info("Using Flash Attention backend on V1 engine.")
198
+ logger.info_once("Using Flash Attention backend on V1 engine.")
199
return ("vllm.v1.attention.backends.flash_attn."
200
"FlashAttentionBackend")
201
if selected_backend == _Backend.FLASHINFER: