From a5d4a264dd2e05d09ab03e7fe113c6cde3096736 Mon Sep 17 00:00:00 2001
From: JiangWeixiang <854746559@qq.com>
Date: Tue, 28 Apr 2026 13:39:23 +0800
Subject: [PATCH] [Bugfix ] fix bailing_moe_linear (#40859)

Signed-off-by: ghphotoframe <854746559@qq.com>
(cherry picked from commit ed57f771923703998a17ad656536ffb460447a2c)
---
 .../layers/mamba/mamba_utils.py               |  3 --
 .../models/bailing_moe_linear.py              | 28 ++++++++++---------
 2 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/vllm/model_executor/layers/mamba/mamba_utils.py b/vllm/model_executor/layers/mamba/mamba_utils.py
index a5a30502b21..c1fd81e40e3 100644
--- a/vllm/model_executor/layers/mamba/mamba_utils.py
+++ b/vllm/model_executor/layers/mamba/mamba_utils.py
@@ -55,9 +55,6 @@ class MambaStateDtypeCalculator:
         model_dtype: ModelDType | torch.dtype,
         mamba_cache_dtype: MambaDType,
     ) -> tuple[torch.dtype, ...]:
-        # TODO (tdoublep) requires testing
-        if mamba_cache_dtype == "float32":
-            raise ValueError("fp32 state for minimax is not yet supported")
         state_dtype = get_kv_cache_torch_dtype(mamba_cache_dtype, model_dtype)
         return (state_dtype,)
 
diff --git a/vllm/model_executor/models/bailing_moe_linear.py b/vllm/model_executor/models/bailing_moe_linear.py
index e26adc17430..55ea1bad44d 100644
--- a/vllm/model_executor/models/bailing_moe_linear.py
+++ b/vllm/model_executor/models/bailing_moe_linear.py
@@ -17,6 +17,7 @@ from vllm.distributed import (
 )
 from vllm.forward_context import get_forward_context
 from vllm.logger import init_logger
+from vllm.model_executor.custom_op import PluggableLayer
 from vllm.model_executor.layers.fla.ops.layernorm_guard import (
     RMSNormGated,
     layernorm_fn,
@@ -211,7 +212,6 @@ class BailingMoeV25MLAAttention(nn.Module):
             max_position=max_position,
             is_neox_style=False,
             rope_parameters=rope_parameters or None,
-            dtype=torch.float32,
         )
 
         # Build MLAModules for MultiHeadLatentAttentionWrapper
@@ -425,14 +425,18 @@ class BailingGroupRMSNormGate(RMSNormGated):
         param.data.copy_(loaded_weight[shard].contiguous())
 
 
-class BailingMoELinearAttention(nn.Module, MambaBase):
-    """
-    Bailing MoE Linear Attention implementation using minimax backend.
+# --8<-- [start:bailing_moe_linear_attention]
+@PluggableLayer.register("bailing_moe_linear_attention")
+class BailingMoELinearAttention(PluggableLayer, MambaBase):
+    """Pluggable Bailing MoE Linear Attention layer which allows OOT backends
+    to add custom implementations.
 
-    This implements the linear attention mechanism from sglang, adapted for vLLM's
-    v1 engine with MambaBase interface support.
+    This implements the linear attention mechanism from sglang, adapted for
+    vLLM's v1 engine with MambaBase interface support.
     """
 
+    # --8<-- [end:bailing_moe_linear_attention]
+
     @property
     def mamba_type(self) -> str:
         return "linear_attention"
@@ -569,7 +573,6 @@ class BailingMoELinearAttention(nn.Module, MambaBase):
             self.head_dim,
             max_position=self.max_position_embeddings,
             is_neox_style=True,
-            dtype=torch.float32,
             rope_parameters=rope_parameters or None,
         )
 
@@ -754,8 +757,6 @@ class BailingMoELinearAttention(nn.Module, MambaBase):
 
     def _decode_infer(self, q, k, v, kv_cache, state_indices_tensor, attn_metadata):
         """Handle decode (single token per sequence)."""
-        num_prefill_tokens = attn_metadata.num_prefill_tokens
-        num_prefills = attn_metadata.num_prefills
         hidden = linear_attention_decode(
             q,
             k,
@@ -763,10 +764,10 @@ class BailingMoELinearAttention(nn.Module, MambaBase):
             kv_cache,
             self.tp_slope,
             state_indices_tensor,
-            q_start=num_prefill_tokens,
-            q_end=None,
-            slot_start=num_prefills,
-            slot_end=None,
+            q_start=0,
+            q_end=attn_metadata.num_decode_tokens,
+            slot_start=0,
+            slot_end=attn_metadata.num_decodes,
             block_size=32,
         )
         return hidden
@@ -1149,6 +1150,7 @@ class BailingMoeV25ForCausalLM(nn.Module, HasInnerState, IsHybrid, SupportsPP):
                 config.vocab_size,
                 config.hidden_size,
                 quant_config=quant_config,
+                prefix=maybe_prefix(prefix, "lm_head"),
             )
             self.logits_processor = LogitsProcessor(config.vocab_size)
         else: