From c9e2090101bc4fc1c84caa0f6c449671a942f532 Mon Sep 17 00:00:00 2001
From: Kevin Li <32230403+bluecoffee8@users.noreply.github.com>
Date: Mon, 1 Dec 2025 21:04:14 -0800
Subject: [PATCH] fix: Support PP for Mistral Small 3.1 (#14254)

---
 python/sglang/srt/models/llava.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/python/sglang/srt/models/llava.py b/python/sglang/srt/models/llava.py
index 2fbbe5590..c76312c0a 100644
--- a/python/sglang/srt/models/llava.py
+++ b/python/sglang/srt/models/llava.py
@@ -41,7 +41,7 @@ from sglang.srt.managers.schedule_batch import (
     MultimodalDataItem,
     MultimodalInputs,
 )
-from sglang.srt.model_executor.forward_batch_info import ForwardBatch
+from sglang.srt.model_executor.forward_batch_info import ForwardBatch, PPProxyTensors
 from sglang.srt.model_loader.weight_utils import default_weight_loader
 from sglang.srt.models.llama import LlamaForCausalLM
 from sglang.srt.models.mistral import MistralForCausalLM
@@ -785,6 +785,7 @@ class LlavaForConditionalGeneration(LlavaBaseForCausalLM):
         positions: torch.Tensor,
         forward_batch: ForwardBatch,
         get_embedding: bool = False,
+        pp_proxy_tensors: Optional[PPProxyTensors] = None,
     ):
         hidden_states = general_mm_embed_routine(
             input_ids=input_ids,
@@ -796,6 +797,7 @@ class LlavaForConditionalGeneration(LlavaBaseForCausalLM):
             },
             placeholder_tokens=None,  # using mm_item.pad_value
             positions=positions,
+            pp_proxy_tensors=pp_proxy_tensors,
         )
 
         return hidden_states