[EP] Enable DBO with NIXL EP (#45275)
Signed-off-by: Itay Alroy <ialroy@nvidia.com>
This commit is contained in:
@@ -1197,7 +1197,7 @@ class CompilationConfig:
|
||||
"are optimized for prefill and are incompatible with CUDA Graphs. "
|
||||
"In order to use CUDA Graphs for decode-optimized workloads, "
|
||||
"use --all2all-backend with another option, such as "
|
||||
"deepep_low_latency or allgather_reducescatter."
|
||||
"deepep_low_latency, nixl_ep, or allgather_reducescatter."
|
||||
)
|
||||
self.cudagraph_mode = CUDAGraphMode.NONE
|
||||
|
||||
|
||||
+7
-5
@@ -1457,12 +1457,14 @@ class VllmConfig:
|
||||
assert a2a_backend in [
|
||||
"deepep_low_latency",
|
||||
"deepep_high_throughput",
|
||||
"nixl_ep",
|
||||
], (
|
||||
"Microbatching currently only supports the deepep_low_latency and "
|
||||
f"deepep_high_throughput all2all backend. {a2a_backend} is not "
|
||||
"supported. To fix use --all2all-backend=deepep_low_latency or "
|
||||
"--all2all-backend=deepep_high_throughput and install the DeepEP"
|
||||
" kernels."
|
||||
"Microbatching currently only supports the deepep_low_latency, "
|
||||
"deepep_high_throughput, and nixl_ep all2all backends. "
|
||||
f"{a2a_backend} is not supported. To fix use "
|
||||
"--all2all-backend=deepep_low_latency, "
|
||||
"--all2all-backend=deepep_high_throughput, or "
|
||||
"--all2all-backend=nixl_ep and install the matching kernels."
|
||||
)
|
||||
|
||||
if not self.model_config.disable_cascade_attn:
|
||||
|
||||
Reference in New Issue
Block a user