qumphy.models.utils.selfattention_family module

class qumphy.models.utils.selfattention_family.AttentionLayer(*args: Any, **kwargs: Any)[source]

Bases: Module

forward(queries, keys, values, attn_mask, tau=None, delta=None)[source]
class qumphy.models.utils.selfattention_family.FlashAttention(*args: Any, **kwargs: Any)[source]

Bases: Module

flash_attention_forward(Q, K, V, mask=None)[source]
forward(queries, keys, values, attn_mask, tau=None, delta=None)[source]
class qumphy.models.utils.selfattention_family.FlowAttention(*args: Any, **kwargs: Any)[source]

Bases: Module

forward(queries, keys, values, attn_mask, tau=None, delta=None)[source]
kernel_method(x)[source]
class qumphy.models.utils.selfattention_family.FullAttention(*args: Any, **kwargs: Any)[source]

Bases: Module

forward(queries, keys, values, attn_mask, tau=None, delta=None)[source]
class qumphy.models.utils.selfattention_family.ProbAttention(*args: Any, **kwargs: Any)[source]

Bases: Module

forward(queries, keys, values, attn_mask, tau=None, delta=None)[source]
class qumphy.models.utils.selfattention_family.ReformerLayer(*args: Any, **kwargs: Any)[source]

Bases: Module

fit_length(queries)[source]
forward(queries, keys, values, attn_mask, tau, delta)[source]