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 15, 06:00 UTC
Fix some typo (#19475)
Signed-off-by: ximing.wxm <ximing.wxm@antgroup.com> Co-authored-by: ximing.wxm <ximing.wxm@antgroup.com>
Ximingwang-09 committed
Jun 11, 2025 at 18:36 UTC
3c8694eabe60e37fbbf2e71aa1414f1370b5014b
1 file changed
+3
-3
vllm/model_executor/layers/rejection_sampler.py
+3
-3
@@ -283,14 +283,14 @@ class RejectionSampler(SpecDecodeStochasticBaseSampler):
283
batch_size, k, _ = draft_probs.shape
284
batch_indices = torch.arange(batch_size,
285
device=target_probs.device)[:, None]
286
- probs_indicies = torch.arange(k, device=target_probs.device)
286
+ probs_indices = torch.arange(k, device=target_probs.device)
287
288
# shape [batch_size, k]
289
- selected_draft_probs = draft_probs[batch_indices, probs_indicies,
289
+ selected_draft_probs = draft_probs[batch_indices, probs_indices,
290
draft_token_ids]
291
292
# shape [batch_size, k]
293
- selected_target_probs = target_probs[batch_indices, probs_indicies,
293
+ selected_target_probs = target_probs[batch_indices, probs_indices,
294
draft_token_ids]
295
296
uniform_rand = self._create_uniform_samples(seeded_seqs, batch_size,