mirror of
https://github.com/vllm-project/vllm.git
synced 2025-10-20 14:53:52 +08:00
[Bugfix] Check chain_speculative_sampling before calling it (#11673)
Signed-off-by: Lu Fang <lufang@fb.com>
This commit is contained in:
@ -118,7 +118,7 @@ class RejectionSampler(SpecDecodeStochasticBaseSampler):
|
||||
|
||||
# If use Flashinfer chain_speculative_sampling kernel
|
||||
# for rejection sampling
|
||||
if self.use_flashinfer:
|
||||
if self.use_flashinfer and chain_speculative_sampling is not None:
|
||||
batch_size, k, _ = draft_probs.shape
|
||||
uniform_samples = self._create_uniform_samples(
|
||||
seeded_seqs, batch_size, k, draft_probs.device)
|
||||
|
Reference in New Issue
Block a user