# Copyright 2025 The EasyDeL/ejKernel Author @erfanzar (Erfan Zare Chavoshi).
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Interface for Flash Multi-Latent Attention (MLA) operations."""
import jaxtyping
from beartype import beartype
from jaxtyping import Array, Float
from ..._registry import Backend, Platform, kernel_registry
[docs]@kernel_registry.register("flash_mla_attention_call", Platform.TRITON, Backend.GPU)
@jaxtyping.jaxtyped(typechecker=beartype)
def flash_mla_attention_call(
query: Float[Array, "batch num_heads seq_len head_dim"],
key: Float[Array, "batch num_heads seq_len head_dim"],
value: Float[Array, "batch num_heads seq_len head_dim"],
latent_key: Float[Array, "head_dim latent_dim"],
latent_value: Float[Array, "head_dim latent_dim"],
bias: Float[Array, "batch num_heads seq_len seq_len"] | None = None,
causal: bool = False,
softmax_scale: float | None = None,
) -> Float[Array, "batch num_heads seq_len head_dim"]:
"""
Execute Multi-Latent Attention using Triton kernels.
Multi-Latent Attention reduces memory and computation by projecting
key and value tensors to lower-dimensional latent spaces before
computing attention.
Args:
query: Query tensor of shape (batch, heads, seq_len, head_dim).
key: Key tensor of shape (batch, heads, seq_len, head_dim).
value: Value tensor of shape (batch, heads, seq_len, head_dim).
latent_key: Latent key projection matrix of shape (head_dim, latent_dim).
latent_value: Latent value projection matrix of shape (head_dim, latent_dim).
bias: Optional attention bias of shape (batch, heads, seq_len, seq_len).
causal: Whether to apply causal masking.
softmax_scale: Scale factor for softmax. Defaults to 1/sqrt(head_dim).
Returns:
Output tensor of shape (batch, heads, seq_len, head_dim).
"""
raise NotImplementedError("Flash MLA attention kernel not yet implemented")
[docs]@kernel_registry.register("flash_mla", Platform.TRITON, Backend.GPU)
@jaxtyping.jaxtyped(typechecker=beartype)
def flash_mla_attention(
query: Float[Array, "batch num_heads seq_len head_dim"],
key: Float[Array, "batch num_heads seq_len head_dim"],
value: Float[Array, "batch num_heads seq_len head_dim"],
latent_key: Float[Array, "head_dim latent_dim"],
latent_value: Float[Array, "head_dim latent_dim"],
bias: Float[Array, "batch num_heads seq_len seq_len"] | None = None,
causal: bool = False,
softmax_scale: float | None = None,
) -> Float[Array, "batch num_heads seq_len head_dim"]:
"""
Multi-Latent Attention with automatic differentiation support.
This function wraps flash_mla_attention_call with JAX's custom gradient
support for efficient backpropagation through the attention operation.
Args:
query: Query tensor of shape (batch, heads, seq_len, head_dim).
key: Key tensor of shape (batch, heads, seq_len, head_dim).
value: Value tensor of shape (batch, heads, seq_len, head_dim).
latent_key: Latent key projection matrix of shape (head_dim, latent_dim).
latent_value: Latent value projection matrix of shape (head_dim, latent_dim).
bias: Optional attention bias of shape (batch, heads, seq_len, seq_len).
causal: Whether to apply causal masking.
softmax_scale: Scale factor for softmax. Defaults to 1/sqrt(head_dim).
Returns:
Output tensor of shape (batch, heads, seq_len, head_dim).
"""
return flash_mla_attention_call(
query=query,
key=key,
value=value,
latent_key=latent_key,
latent_value=latent_value,
bias=bias,
causal=causal,
softmax_scale=softmax_scale,
)