issue/143 init input to support warmup

csrc/engine/compiler/paged_compiler.cpp

@@ -33,7 +33,14 @@ void PagedCompiler::compile() {
             input.input_ids = infinicore::Tensor::empty({1, b}, infinicore::DataType::I64, infinicore::context::getDevice());
             input.position_ids = infinicore::Tensor::empty({b}, infinicore::DataType::I64, infinicore::context::getDevice());
             input.total_sequence_lengths = infinicore::Tensor::empty({b}, infinicore::DataType::I64, infinicore::context::getDevice());
+            std::vector<int64_t> total_sequence_lengths_vec(b, 1);
+            infinicore::context::memcpyH2D(input.total_sequence_lengths.value()->data(), total_sequence_lengths_vec.data(), b * sizeof(int64_t), false);
             input.input_offsets = infinicore::Tensor::empty({b + 1}, infinicore::DataType::I64, infinicore::context::getDevice());
+            std::vector<int64_t> input_offsets_vec(b + 1, 0);
+            for (size_t i = 0; i <= b; i++) {
+                input_offsets_vec[i] = i;
+            }
+            infinicore::context::memcpyH2D(input.input_offsets.value()->data(), input_offsets_vec.data(), (b + 1) * sizeof(int64_t), false);
             input.block_tables = block_tables_holder_->as_strided({b, block_per_req}, {(ptrdiff_t)block_per_req, 1});
             input.slot_mapping = infinicore::Tensor::empty({b}, infinicore::DataType::I64, infinicore::context::getDevice());
             infinicore::context::startGraphRecording();

csrc/engine/compiler/static_batching_compiler.cpp

@@ -15,6 +15,8 @@ void StaticBatchingCompiler::compile() {
         input.position_ids = infinicore::Tensor::empty({b, 1}, infinicore::DataType::I64, infinicore::context::getDevice());
         input.past_sequence_lengths = infinicore::Tensor::empty({b}, infinicore::DataType::I64, infinicore::context::getDevice());
         input.total_sequence_lengths = infinicore::Tensor::empty({b}, infinicore::DataType::I64, infinicore::context::getDevice());
+        std::vector<int64_t> total_sequence_lengths_vec(b, 1);
+        infinicore::context::memcpyH2D(input.total_sequence_lengths.value()->data(), total_sequence_lengths_vec.data(), b * sizeof(int64_t), false);
         infinicore::context::startGraphRecording();
         auto output = model_->forward(input);
         auto graph = infinicore::context::stopGraphRecording();