hotfix: revert sampler CUDA Graph (#1242)

python/sglang/srt/layers/logits_processor.py

@@ -29,7 +29,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardMode, InputMetad
 
 
 @dataclasses.dataclass
-class LogitsProcessorOutput:
+class LogitProcessorOutput:
     # The logits of the next tokens.       shape: [#seq, vocab_size]
     next_token_logits: torch.Tensor
     # The logprobs of the next tokens.     shape: [#seq, vocab_size]
@@ -185,7 +185,7 @@ class LogitsProcessor(nn.Module):
 
         # Return only last_logits if logprob is not requested
         if not logits_metadata.return_logprob:
-            return LogitsProcessorOutput(
+            return LogitProcessorOutput(
                 next_token_logits=last_logits,
                 next_token_logprobs=None,
                 normalized_prompt_logprobs=None,
@@ -209,7 +209,7 @@ class LogitsProcessor(nn.Module):
                 else:
                     output_top_logprobs = None
 
-                return LogitsProcessorOutput(
+                return LogitProcessorOutput(
                     next_token_logits=last_logits,
                     next_token_logprobs=last_logprobs,
                     normalized_prompt_logprobs=None,
@@ -278,7 +278,7 @@ class LogitsProcessor(nn.Module):
                 # Remove the last token logprob for the prefill tokens.
                 input_token_logprobs = input_token_logprobs[:-1]
 
-                return LogitsProcessorOutput(
+                return LogitProcessorOutput(
                     next_token_logits=last_logits,
                     next_token_logprobs=last_logprobs,
                     normalized_prompt_logprobs=normalized_prompt_logprobs,