[Model] Adding support for MiniCPM-V (#4087)

2024-07-25 11:59:30 +08:00
parent 5689e256ba
commit 9e169a4c61
11 changed files with 942 additions and 18 deletions
--- a/vllm/model_executor/models/minicpm.py
+++ b/vllm/model_executor/models/minicpm.py
@@ -463,10 +463,11 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA):
        positions: torch.Tensor,
        kv_caches: List[torch.Tensor],
        attn_metadata: AttentionMetadata,
+        input_embeds: Optional[torch.Tensor] = None,
        intermediate_tensors: Optional[IntermediateTensors] = None,
    ) -> torch.Tensor:
        hidden_states = self.model(input_ids, positions, kv_caches,
-                                   attn_metadata)
+                                   attn_metadata, input_embeds)
        return hidden_states

    def compute_logits(self, hidden_states: torch.Tensor,