bugfix: fix min-p AOT compilation in #713 (#717)

#713 didn't update the function signature of min-p API in AOT pybind file, this PR fixes the issue.
flashinfer-ai · Jan 4, 2025 · def019e · def019e
1 parent 989dbfa
commit def019e
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/csrc/flashinfer_ops.cu b/csrc/flashinfer_ops.cu
@@ -174,8 +174,8 @@ void top_k_sampling_from_probs(at::Tensor probs, at::Tensor uniform_samples, at:
                                unsigned int top_k_val, bool deterministic, int64_t cuda_stream);
 
 void min_p_sampling_from_probs(at::Tensor probs, at::Tensor uniform_samples, at::Tensor samples,
-                               at::Tensor success, std::optional<at::Tensor> maybe_min_p_arr,
-                               double min_p_val, bool deterministic, int64_t cuda_stream);
+                               std::optional<at::Tensor> maybe_min_p_arr, double min_p_val,
+                               bool deterministic, int64_t cuda_stream);
 
 void top_k_top_p_sampling_from_probs(at::Tensor probs, at::Tensor uniform_samples,
                                      at::Tensor samples, at::Tensor success,