[SpecDecode]Fix spec decode warmup device selection (#55245)

Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
Co-authored-by: GitHub Copilot <copilot@github.com>
This commit is contained in:
Kunshang Ji
2026-09-04 00:48:45 -07:00
committed by GitHub
co-authored by GitHub Copilot
parent a5c9179e73
commit a8693df504
@@ -44,7 +44,7 @@ def spec_decode_rejection_warmup(worker: Worker) -> None:
use_block_verification = rejection_method == "block"
use_synthetic = rejection_method == "synthetic"
device = torch.device("cuda")
device = worker.device
num_reqs = 1
tokens_per_req = num_spec + 1
num_logits = num_reqs * tokens_per_req