Deepseekv32 compatibility with transformers v5 (#18297)

Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>

python/sglang/srt/layers/attention/nsa_backend.py

@@ -300,6 +300,8 @@ class NativeSparseAttnBackend(
         self.req_to_token = model_runner.req_to_token_pool.req_to_token
 
         self.use_mha: bool = False
+        # Force NSA prefill to use MLA (i.e. disable MHA_ONE_SHOT), controlled by env var.
+        self._force_attn_forward_mla: bool = envs.SGLANG_NSA_FORCE_MLA.get()
         self.nsa_prefill_impl: _NSA_IMPL_T = (
             model_runner.server_args.nsa_prefill_backend
         )
@@ -1837,6 +1839,8 @@ class NativeSparseAttnBackend(
             )
         else:
             self.use_mha = False  # Decode/verify always use MLA
+        if self._force_attn_forward_mla:
+            self.use_mha = False
 
         # Set MLA implementation only if not using MHA
         if not self.use_mha and self.enable_auto_select_prefill_impl: