[Fix] Fix serve model to adapt the latest Allocator signature (#1989)

MasterJH5574 · web-flow · commit d4ec25edb280 · 2024-03-20T15:42:18.000-04:00
PR apache/tvm#16738 updated the Allocator signature. This PR updates the caller side accordingly.
diff --git a/cpp/serve/model.cc b/cpp/serve/model.cc
@@ -64,7 +64,7 @@ class ModelImpl : public ModelObj {
         memory::MemoryManager::GetOrCreateAllocator(device_host, memory::AllocatorType::kNaive);
     ICHECK_NOTNULL(allocator);
     token_ids_storage_ =
-        memory::Storage(allocator->Alloc({prefill_chunk_size_}, DataType::Int(32)));
+        memory::Storage(allocator->Alloc(device_host, {prefill_chunk_size_}, DataType::Int(32)));
     this->logit_pos_arr_ = NDArray::Empty({max_num_sequence}, DataType::Int(32), device_host);
   }
 

Original file line number	Diff line number	Diff line change
`@@ -64,7 +64,7 @@ class ModelImpl : public ModelObj {`
`64`	`64`	`memory::MemoryManager::GetOrCreateAllocator(device_host, memory::AllocatorType::kNaive);`
`65`	`65`	`ICHECK_NOTNULL(allocator);`
`66`	`66`	`token_ids_storage_ =`
`67`		`- memory::Storage(allocator->Alloc({prefill_chunk_size_}, DataType::Int(32)));`
	`67`	`+ memory::Storage(allocator->Alloc(device_host, {prefill_chunk_size_}, DataType::Int(32)));`
`68`	`68`	`this->logit_pos_arr_ = NDArray::Empty({max_num_sequence}, DataType::Int(32), device_host);`
`69`	`69`	`}`
`70`	`70`