Return a mamba tracking entry from the cache lookup instead of mutating caller lists (#25724)

python/sglang/srt/managers/schedule_batch.py

@@ -40,7 +40,17 @@ from enum import Enum, auto
 from functools import lru_cache
 from http import HTTPStatus
 from itertools import chain
-from typing import TYPE_CHECKING, Any, Dict, List, Optional, Set, Tuple, Union
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    Dict,
+    List,
+    NamedTuple,
+    Optional,
+    Set,
+    Tuple,
+    Union,
+)
 
 import numpy as np
 import torch
@@ -1379,6 +1389,12 @@ class Req(ReqDllmMixin):
         )
 
 
+class _MambaRadixCacheV2TrackEntry(NamedTuple):
+    track_mask: bool
+    track_index: int
+    track_seqlen: int
+
+
 @dataclasses.dataclass
 class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
     """Store all information of a batch on the scheduler."""
@@ -1850,12 +1866,10 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
             req.is_retracted = False
 
             if get_global_server_args().enable_mamba_extra_buffer():
-                self._mamba_radix_cache_v2_req_prepare_for_extend(
+                track_entry = self._mamba_radix_cache_v2_req_prepare_for_extend(req)
-                    req,
+                mamba_track_mask_cpu.append(track_entry.track_mask)
-                    mamba_track_mask_cpu,
+                mamba_track_indices_cpu.append(track_entry.track_index)
-                    mamba_track_indices_cpu,
+                mamba_track_seqlens_cpu.append(track_entry.track_seqlen)
-                    mamba_track_seqlens_cpu,
-                )
 
             if self.return_logprob:
                 # Find input logprob token ids.
@@ -2001,10 +2015,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
     def _mamba_radix_cache_v2_req_prepare_for_extend(
         self,
         req: Req,
-        mamba_track_mask_cpu: List[bool],
+    ) -> "_MambaRadixCacheV2TrackEntry":
-        mamba_track_indices_cpu: List[int],
-        mamba_track_seqlens_cpu: List[int],
-    ):
         def _force_track_h(i: int) -> int:
             assert i % FLA_CHUNK_SIZE == 0
             # There are 3 cases for mamba_track_seqlen passed to mamba_track_seqlens_cpu:
@@ -2018,10 +2029,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
 
         mamba_cache_chunk_size = get_global_server_args().mamba_cache_chunk_size
         mask = req.extend_input_len >= mamba_cache_chunk_size
-        mamba_track_mask_cpu.append(mask)
+        track_index = req.mamba_ping_pong_track_buffer[req.mamba_next_track_idx].item()
-        mamba_track_indices_cpu.append(
-            req.mamba_ping_pong_track_buffer[req.mamba_next_track_idx].item()
-        )
         mamba_track_seqlen = -1
         if mask:
             # mamba_track_seqlen is used to calculate the indices to track in
@@ -2076,7 +2084,11 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
                     mamba_track_seqlen = _force_track_h(req.mamba_branching_seqlen)
                     mamba_track_seqlen_aligned = req.mamba_branching_seqlen
             req.mamba_last_track_seqlen = mamba_track_seqlen_aligned
-        mamba_track_seqlens_cpu.append(mamba_track_seqlen)
+        return _MambaRadixCacheV2TrackEntry(
+            track_mask=mask,
+            track_index=track_index,
+            track_seqlen=mamba_track_seqlen,
+        )
 
     def mix_with_running(self, running_batch: "ScheduleBatch"):
         self.forward_mode = ForwardMode.MIXED