[Feature] Add MiniCPM-SALA support (#30360)

Co-authored-by: Alex Nails <alex.nails@radixark.ai>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

python/sglang/srt/model_executor/model_runner.py

@@ -1298,9 +1298,12 @@ class ModelRunner:
     def effective_max_total_num_tokens(self):
         """Return the max token pool size considering hybrid swa settings."""
         if self.is_hybrid_swa:
-            return self.full_max_total_num_tokens or self.swa_max_total_num_tokens
+            capacity = self.full_max_total_num_tokens or self.swa_max_total_num_tokens
         else:
-            return self.max_total_num_tokens
+            capacity = self.max_total_num_tokens
+        if (req_to_token_pool := getattr(self, "req_to_token_pool", None)) is not None:
+            return req_to_token_pool.schedulable_token_capacity(capacity)
+        return capacity
 
     @property
     def max_token_pool_size(self):

python/sglang/srt/model_executor/model_runner_components/kv_pool_runtime.py

@@ -96,6 +96,7 @@ def compute_post_capture_kv_resize(
     )
     pool.finalize_backing(config)
     model_runner.token_to_kv_pool_allocator.resize(config)
+    model_runner.req_to_token_pool.reset_aux_cache_allocator()
 
     capped_max_running_requests = None
     if model_runner.max_running_requests is not None: