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 4, 08:45 UTC
Disable flashinfer autotune temporarily due to correctness issues (#41524)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Wei Zhao committed
May 3, 2026 at 12:19 UTC
c51df43005726a09c6eb7348e8c1b00501c70a8e
1 file changed
+6
-2
vllm/config/vllm.py
+6
-2
@@ -209,7 +209,9 @@ OPTIMIZATION_LEVEL_01 = {
209
"use_inductor_graph_partition": False,
210
},
211
"kernel_config": {
212
- "enable_flashinfer_autotune": True,
212
+ # Disabled for now due to correctness issues:
213
+ # https://github.com/flashinfer-ai/flashinfer/issues/3197
214
+ "enable_flashinfer_autotune": False,
215
},
216
}
217
OPTIMIZATION_LEVEL_02 = {
@@ -229,7 +231,9 @@ OPTIMIZATION_LEVEL_02 = {
231
"use_inductor_graph_partition": False,
232
},
233
"kernel_config": {
232
- "enable_flashinfer_autotune": True,
234
+ # Disabled for now due to correctness issues:
235
+ # https://github.com/flashinfer-ai/flashinfer/issues/3197
236
+ "enable_flashinfer_autotune": False,
237
},
238
}
239
OPTIMIZATION_LEVEL_03 = {