Merge branch 'v0.9.2-dev-wm' into 'v0.9.2-dev'

[fix]v0 SamplerOutput在非tree decoding时不传入logits

See merge request dcutoolkit/deeplearing/vllm!185

vllm/model_executor/layers/sampler.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """A layer that samples the next tokens from the model's outputs."""
+import os
 import itertools
 from collections.abc import Iterator
 from dataclasses import dataclass
@@ -210,6 +211,7 @@ class Sampler(nn.Module):
         # speculative decoding and when prompt embeddings are specified.
         self.include_gpu_probs_tensor = False
         self.should_modify_greedy_probs_inplace = False
+        self.tree_decoding = (os.environ.get('VLLM_TREE_DECODING') == '1')
 
     def _init_sampling_tensors(
         self,
@@ -347,7 +349,7 @@ class Sampler(nn.Module):
             sample_logprobs,
             on_device_tensors=on_device_tensors,
             skip_sampler_cpu_output=sampling_metadata.skip_sampler_cpu_output,
-            logits=logits)
+            logits=logits if self.tree_decoding else None)
 
     @property
     def _should_modify_greedy_probs_inplace(self) -> bool: