|
Mila
Deep Neural Network Library
|
Minimum contract for all KV cache management policies. More...
Minimum contract for all KV cache management policies.
Requires only kIsActive. Active policies set kIsActive = true and provide algorithm-specific fields consumed by CudaGqaOp dispatch via if constexpr. Active policy refinements such as QuantKvPolicy impose additional field requirements in their own modules. A SlidingWindowPolicy satisfies KvCachePolicy without carrying dtype fields.
| T | The type to constrain against the policy contract. |