[Fix]: support deepseek-vl2-tiny model (#5552)

Co-authored-by: bppps <zouyu.zzx@alibaba-inc.com>

python/sglang/srt/models/deepseek.py

@@ -382,8 +382,14 @@ class DeepseekModel(nn.Module):
         input_ids: torch.Tensor,
         positions: torch.Tensor,
         forward_batch: ForwardBatch,
+        input_embeds: torch.Tensor = None,
     ) -> torch.Tensor:
-        hidden_states = self.embed_tokens(input_ids)
+
+        if input_embeds is None:
+            hidden_states = self.embed_tokens(input_ids)
+        else:
+            hidden_states = input_embeds
+
         residual = None
         for i in range(len(self.layers)):
             layer = self.layers[i]
@@ -416,14 +422,18 @@ class DeepseekForCausalLM(nn.Module):
         )
         self.logits_processor = LogitsProcessor(config)
 
+    def get_input_embeddings(self) -> nn.Embedding:
+        return self.model.embed_tokens
+
     @torch.no_grad()
     def forward(
         self,
         input_ids: torch.Tensor,
         positions: torch.Tensor,
         forward_batch: ForwardBatch,
+        input_embeds: torch.Tensor = None,
     ) -> torch.Tensor:
-        hidden_states = self.model(input_ids, positions, forward_batch)
+        hidden_states = self.model(input_ids, positions, forward_batch, input_embeds)
         return self.logits_processor(
             input_ids, hidden_states, self.lm_head, forward_batch
         )