[RL] Fix FlashInfer TRTLLM MXFP8 dense weight layout (#28459)
This commit is contained in:
@@ -171,7 +171,7 @@ class FlashinferTrtllmGenMoeBackendMXFP8MixedBF16Base:
|
||||
"--kv-cache-dtype",
|
||||
"bf16",
|
||||
"--fp8-gemm-backend",
|
||||
"flashinfer_cutlass",
|
||||
"flashinfer_trtllm",
|
||||
"--moe-runner-backend",
|
||||
cls.backend,
|
||||
"--tp-size",
|
||||
|
||||
Reference in New Issue
Block a user