[Easy] Eliminate c10::optional usage in vllm/csrc (vllm-project#17819)

Signed-off-by: Mu Huai <tianbowen.tbw@antgroup.com>

Author: houseroad

csrc/quantization/gptq_allspark/allspark_qgemm_w8a16.cu

@@ -9,7 +9,7 @@ at::Tensor as_g_workspace;
 
 torch::Tensor allspark_w8a16_gemm(
     torch::Tensor const& a, torch::Tensor const& b_qweight,
-    torch::Tensor const& b_scales, c10::optional<torch::Tensor> const& b_qzeros,
+    torch::Tensor const& b_scales, std::optional<torch::Tensor> const& b_qzeros,
     int64_t n, int64_t group_size, int64_t sm_count, int64_t sm_version,
     int64_t CUBLAS_M_THRESHOLD, bool has_zp, bool n32k16_reorder) {
   TORCH_CHECK_NOT_IMPLEMENTED(
@@ -918,7 +918,7 @@ void allspark_qgemm_w8a16_perc_ampere(
 
 torch::Tensor allspark_w8a16_gemm(
     torch::Tensor const& a, torch::Tensor const& b_qweight,
-    torch::Tensor const& b_scales, c10::optional<torch::Tensor> const& b_qzeros,
+    torch::Tensor const& b_scales, std::optional<torch::Tensor> const& b_qzeros,
     int64_t n, int64_t group_size, int64_t sm_count, int64_t sm_version,
     int64_t CUBLAS_M_THRESHOLD, bool has_zp, bool n32k16_reorder) {
   // Verify device and strides

csrc/quantization/gptq_allspark/allspark_repack.cu

@@ -100,9 +100,9 @@ void rearrange_kn_weight_as_n32k16_order_ldg16(
 
 void rearrange_kn_weight_as_n32k16_order(
     torch::Tensor const& b_qweight, torch::Tensor const& b_scales,
-    c10::optional<torch::Tensor> const& b_zeros, bool has_zp,
+    std::optional<torch::Tensor> const& b_zeros, bool has_zp,
     torch::Tensor& b_qweight_reorder, torch::Tensor& b_scales_reorder,
-    c10::optional<torch::Tensor> const& b_zeros_reorder, const int64_t K,
+    std::optional<torch::Tensor> const& b_zeros_reorder, const int64_t K,
     const int64_t N, const int64_t N_32align) {
   // Verify device and strides
   TORCH_CHECK(b_qweight.device().is_cuda(), "b_qweight is not on GPU");

csrc/rocm/attention.cu

@@ -1597,7 +1597,7 @@ void paged_attention_custom_launcher(
     torch::Tensor& block_tables, torch::Tensor& context_lens,
     const std::optional<torch::Tensor>& query_start_loc, int max_context_len,
     const std::optional<torch::Tensor>& alibi_slopes, torch::Tensor& k_scale,
-    torch::Tensor& v_scale, const c10::optional<torch::Tensor>& fp8_out_scale) {
+    torch::Tensor& v_scale, const std::optional<torch::Tensor>& fp8_out_scale) {
   int num_seqs = block_tables.size(0);
   int num_heads = query.size(1);
   int head_size = query.size(2);
@@ -1825,7 +1825,7 @@ void paged_attention(
     const std::optional<torch::Tensor>& alibi_slopes,
     const std::string& kv_cache_dtype, torch::Tensor& k_scale,
     torch::Tensor& v_scale,
-    const c10::optional<torch::Tensor>& fp8_out_scale) {
+    const std::optional<torch::Tensor>& fp8_out_scale) {
   // clang-format on
   const int head_size = query.size(2);
   if (kv_cache_dtype == "auto") {

csrc/rocm/ops.h

@@ -19,4 +19,4 @@ void paged_attention(
     const std::optional<torch::Tensor>& query_start_loc, int64_t block_size,
     int64_t max_context_len, const std::optional<torch::Tensor>& alibi_slopes,
     const std::string& kv_cache_dtype, torch::Tensor& k_scale,
-    torch::Tensor& v_scale, const c10::optional<torch::Tensor>& fp8_out_scale);
+    torch::Tensor& v_scale, const std::optional<torch::Tensor>& fp8_out_scale);