Revert "Add attn sink (#2892)" (#3250)

This reverts commit 9fa4e8d5ab.
This commit is contained in:
asleepzzz
2025-11-20 23:55:15 +08:00
committed by GitHub
parent 9fa4e8d5ab
commit 5adaa201ed
25 changed files with 195 additions and 940 deletions

View File

@@ -162,17 +162,6 @@ struct StandardAttention
{
return !params.impl_mask.IsOutOfBound(qo_idx, kv_idx);
}
template <typename Params>
__device__ __forceinline__ bool LogitsSinkMask(const Params& params,
[[maybe_unused]] uint32_t batch_idx,
uint32_t qo_idx,
uint32_t kv_idx,
[[maybe_unused]] uint32_t qo_head_idx,
[[maybe_unused]] uint32_t kv_head_idx) const
{
return !params.impl_mask.IsOutOfSinkBound(qo_idx, kv_idx);
}
};
template <bool UseExp2 = false>
@@ -235,17 +224,6 @@ struct LogitsSoftCap
{
return !params.impl_mask.IsOutOfBound(qo_idx, kv_idx);
}
template <typename Params>
__device__ __forceinline__ bool LogitsSinkMask(const Params& params,
[[maybe_unused]] uint32_t batch_idx,
uint32_t qo_idx,
uint32_t kv_idx,
[[maybe_unused]] uint32_t qo_head_idx,
[[maybe_unused]] uint32_t kv_head_idx) const
{
return !params.impl_mask.IsOutOfSinkBound(qo_idx, kv_idx);
}
};
constexpr uint32_t CUSTOM_MASK = 1U;
@@ -319,17 +297,6 @@ struct ComposedAttention
{
return !params.impl_mask.IsOutOfBound(qo_idx, kv_idx);
}
template <typename Params>
__device__ __forceinline__ bool LogitsSinkMask(const Params& params,
[[maybe_unused]] uint32_t batch_idx,
uint32_t qo_idx,
uint32_t kv_idx,
[[maybe_unused]] uint32_t qo_head_idx,
[[maybe_unused]] uint32_t kv_head_idx) const
{
return !params.impl_mask.IsOutOfSinkBound(qo_idx, kv_idx);
}
};
} // namespace ck_tile