#if !defined(TORCH_STABLE_ONLY) && !defined(TORCH_TARGET_VERSION)
#pragma once

// @generated by torchgen/gen.py from Function.h

#include <ATen/Context.h>
#include <ATen/DeviceGuard.h>
#include <ATen/TensorUtils.h>
#include <ATen/TracerMode.h>
#include <ATen/core/Generator.h>
#include <ATen/core/Reduction.h>
#include <ATen/core/Tensor.h>
#include <c10/core/Scalar.h>
#include <c10/core/Storage.h>
#include <c10/core/TensorOptions.h>
#include <c10/util/Deprecated.h>
#include <optional>
#include <string_view>



#include <ATen/ops/_flash_attention_forward_ops.h>

namespace at {


// aten::_flash_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? cum_seq_q, Tensor? cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, bool return_debug_mask, *, float? scale=None, SymInt? window_size_left=None, SymInt? window_size_right=None, Tensor? seqused_k=None, Tensor? alibi_slopes=None, Tensor? block_table=None, int? num_splits=None) -> (Tensor output, Tensor softmax_logsumexp, Tensor rng_state, Tensor unused, Tensor debug_attn_mask)
inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, bool is_causal, bool return_debug_mask, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> window_size_left=::std::nullopt, ::std::optional<int64_t> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}, const ::std::optional<at::Tensor> & block_table={}, ::std::optional<int64_t> num_splits=::std::nullopt) {
    return at::_ops::_flash_attention_forward::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, scale, window_size_left.has_value() ? ::std::make_optional(c10::SymInt(*window_size_left)) : ::std::nullopt, window_size_right.has_value() ? ::std::make_optional(c10::SymInt(*window_size_right)) : ::std::nullopt, seqused_k, alibi_slopes, block_table, num_splits);
}
namespace symint {
  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, bool is_causal, bool return_debug_mask, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> window_size_left=::std::nullopt, ::std::optional<int64_t> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}, const ::std::optional<at::Tensor> & block_table={}, ::std::optional<int64_t> num_splits=::std::nullopt) {
    return at::_ops::_flash_attention_forward::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, scale, window_size_left.has_value() ? ::std::make_optional(c10::SymInt(*window_size_left)) : ::std::nullopt, window_size_right.has_value() ? ::std::make_optional(c10::SymInt(*window_size_right)) : ::std::nullopt, seqused_k, alibi_slopes, block_table, num_splits);
  }
}

// aten::_flash_attention_forward(Tensor query, Tensor key, Tensor value, Tensor? cum_seq_q, Tensor? cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, bool return_debug_mask, *, float? scale=None, SymInt? window_size_left=None, SymInt? window_size_right=None, Tensor? seqused_k=None, Tensor? alibi_slopes=None, Tensor? block_table=None, int? num_splits=None) -> (Tensor output, Tensor softmax_logsumexp, Tensor rng_state, Tensor unused, Tensor debug_attn_mask)
inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward_symint(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, c10::SymInt max_q, c10::SymInt max_k, double dropout_p, bool is_causal, bool return_debug_mask, ::std::optional<double> scale=::std::nullopt, ::std::optional<c10::SymInt> window_size_left=::std::nullopt, ::std::optional<c10::SymInt> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}, const ::std::optional<at::Tensor> & block_table={}, ::std::optional<int64_t> num_splits=::std::nullopt) {
    return at::_ops::_flash_attention_forward::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, scale, window_size_left, window_size_right, seqused_k, alibi_slopes, block_table, num_splits);
}
namespace symint {
  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, c10::SymInt max_q, c10::SymInt max_k, double dropout_p, bool is_causal, bool return_debug_mask, ::std::optional<double> scale=::std::nullopt, ::std::optional<c10::SymInt> window_size_left=::std::nullopt, ::std::optional<c10::SymInt> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}, const ::std::optional<at::Tensor> & block_table={}, ::std::optional<int64_t> num_splits=::std::nullopt) {
    return at::_ops::_flash_attention_forward::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, scale, window_size_left, window_size_right, seqused_k, alibi_slopes, block_table, num_splits);
  }
}

// aten::_flash_attention_forward.quantized(Tensor query, Tensor key, Tensor value, Tensor? cum_seq_q, Tensor? cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, bool return_debug_mask, Tensor? q_descale, Tensor? k_descale, Tensor? v_descale, *, float? scale=None, SymInt? window_size_left=None, SymInt? window_size_right=None, Tensor? seqused_k=None, Tensor? alibi_slopes=None) -> (Tensor output, Tensor softmax_logsumexp, Tensor rng_state, Tensor unused, Tensor debug_attn_mask)
inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, bool is_causal, bool return_debug_mask, const ::std::optional<at::Tensor> & q_descale, const ::std::optional<at::Tensor> & k_descale, const ::std::optional<at::Tensor> & v_descale, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> window_size_left=::std::nullopt, ::std::optional<int64_t> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}) {
    return at::_ops::_flash_attention_forward_quantized::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, q_descale, k_descale, v_descale, scale, window_size_left.has_value() ? ::std::make_optional(c10::SymInt(*window_size_left)) : ::std::nullopt, window_size_right.has_value() ? ::std::make_optional(c10::SymInt(*window_size_right)) : ::std::nullopt, seqused_k, alibi_slopes);
}
namespace symint {
  template <typename T, typename = std::enable_if_t<std::is_same_v<T, int64_t>>>
  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, int64_t max_q, int64_t max_k, double dropout_p, bool is_causal, bool return_debug_mask, const ::std::optional<at::Tensor> & q_descale, const ::std::optional<at::Tensor> & k_descale, const ::std::optional<at::Tensor> & v_descale, ::std::optional<double> scale=::std::nullopt, ::std::optional<int64_t> window_size_left=::std::nullopt, ::std::optional<int64_t> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}) {
    return at::_ops::_flash_attention_forward_quantized::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, q_descale, k_descale, v_descale, scale, window_size_left.has_value() ? ::std::make_optional(c10::SymInt(*window_size_left)) : ::std::nullopt, window_size_right.has_value() ? ::std::make_optional(c10::SymInt(*window_size_right)) : ::std::nullopt, seqused_k, alibi_slopes);
  }
}

// aten::_flash_attention_forward.quantized(Tensor query, Tensor key, Tensor value, Tensor? cum_seq_q, Tensor? cum_seq_k, SymInt max_q, SymInt max_k, float dropout_p, bool is_causal, bool return_debug_mask, Tensor? q_descale, Tensor? k_descale, Tensor? v_descale, *, float? scale=None, SymInt? window_size_left=None, SymInt? window_size_right=None, Tensor? seqused_k=None, Tensor? alibi_slopes=None) -> (Tensor output, Tensor softmax_logsumexp, Tensor rng_state, Tensor unused, Tensor debug_attn_mask)
inline ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward_symint(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, c10::SymInt max_q, c10::SymInt max_k, double dropout_p, bool is_causal, bool return_debug_mask, const ::std::optional<at::Tensor> & q_descale, const ::std::optional<at::Tensor> & k_descale, const ::std::optional<at::Tensor> & v_descale, ::std::optional<double> scale=::std::nullopt, ::std::optional<c10::SymInt> window_size_left=::std::nullopt, ::std::optional<c10::SymInt> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}) {
    return at::_ops::_flash_attention_forward_quantized::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, q_descale, k_descale, v_descale, scale, window_size_left, window_size_right, seqused_k, alibi_slopes);
}
namespace symint {
  template <typename T, typename = std::enable_if_t<std::is_same_v<T, c10::SymInt>>>
  ::std::tuple<at::Tensor,at::Tensor,at::Tensor,at::Tensor,at::Tensor> _flash_attention_forward(const at::Tensor & query, const at::Tensor & key, const at::Tensor & value, const ::std::optional<at::Tensor> & cum_seq_q, const ::std::optional<at::Tensor> & cum_seq_k, c10::SymInt max_q, c10::SymInt max_k, double dropout_p, bool is_causal, bool return_debug_mask, const ::std::optional<at::Tensor> & q_descale, const ::std::optional<at::Tensor> & k_descale, const ::std::optional<at::Tensor> & v_descale, ::std::optional<double> scale=::std::nullopt, ::std::optional<c10::SymInt> window_size_left=::std::nullopt, ::std::optional<c10::SymInt> window_size_right=::std::nullopt, const ::std::optional<at::Tensor> & seqused_k={}, const ::std::optional<at::Tensor> & alibi_slopes={}) {
    return at::_ops::_flash_attention_forward_quantized::call(query, key, value, cum_seq_q, cum_seq_k, max_q, max_k, dropout_p, is_causal, return_debug_mask, q_descale, k_descale, v_descale, scale, window_size_left, window_size_right, seqused_k, alibi_slopes);
  }
}

}

#else
#error "This file should not be included when either TORCH_STABLE_ONLY or TORCH_TARGET_VERSION is defined."
#endif  // !defined(TORCH_STABLE_ONLY) && !defined(TORCH_TARGET_VERSION)