Fix Illegal Instruction/IMA errors when using DP attention --...

Fix Illegal Instruction/IMA errors when using DP attention -- num_tokens_for_logprob calculation (#12115)

python/sglang/srt/managers/scheduler.py

@@ -2073,15 +2073,18 @@ class Scheduler(
             num_tokens_for_logprob = num_tokens
         else:
             num_tokens = local_batch.extend_num_tokens
+            if local_batch.return_logprob:
                 num_tokens_for_logprob = sum(
-                [
                     # We should have at least 1 token for sample in every case.
                     max(extend_len - logprob_start_len, 1)
                     for logprob_start_len, extend_len in zip(
-                        local_batch.extend_logprob_start_lens, local_batch.extend_lens
+                        local_batch.extend_logprob_start_lens,
+                        local_batch.extend_lens,
                     )
-                ]
                 )
+            else:
+                # When return_logprob = False, only need last token per request
+                num_tokens_for_logprob = local_batch.batch_size()
 
         if local_batch is None or local_batch.forward_mode.is_decode_or_idle():
             can_cuda_graph = 1