MCPcopy Create free account

hub / github.com/fla-org/flash-linear-attention / functions

Functions2,989 in github.com/fla-org/flash-linear-attention

Method_init_weights
( self, module: nn.Module, num_residuals_per_layer: int = 1, )
fla/models/mamba3/modeling_mamba3.py:97
Method_init_weights
( self, module: nn.Module, prenorm_residual_strategy: str | None = None, num_r
fla/models/lightnet/modeling_lightnet.py:180
Method_init_weights
( self, module: nn.Module, rescale_prenorm_residual: bool = False, num_residua
fla/models/transformer/modeling_transformer.py:183
Method_init_weights
( self, module: nn.Module, prenorm_residual_strategy: str | None = None, num_r
fla/models/comba/modeling_comba.py:183
Method_init_weights
( self, module: nn.Module, prenorm_residual_strategy: str | None = None, num_r
fla/models/nsa/modeling_nsa.py:183
Method_init_weights
( self, module: nn.Module, prenorm_residual_strategy: str | None = None, num_r
fla/models/kda/modeling_kda.py:196
Method_init_weights
( self, module: nn.Module, prenorm_residual_strategy: str | None = None, num_r
fla/models/gsa/modeling_gsa.py:189
Method_initialize_weights
(self, module: nn.Module)
fla/layers/rwkv7.py:157
Method_initialize_weights
(self, module: nn.Module)
fla/layers/mom.py:416
Method_initialize_weights
(self, module: nn.Module)
fla/layers/gated_deltaproduct.py:155
Method_initialize_weights
(self, module: nn.Module)
fla/layers/rwkv6.py:103
Method_initialize_weights
(self, module: nn.Module)
fla/layers/rwkv6.py:259
Method_initialize_weights
(self, module: nn.Module)
fla/models/rwkv7/modeling_rwkv7.py:84
Function_mamba3_decode_kernels_available
Decode requires the cute step kernel and the rotary step kernel, which are independent of the Triton prefill kernel.
tests/models/test_modeling_mamba3.py:23
Function_official_mamba2_step_fixed_d_has_hdim
Copy of mamba_ssm Mamba2.step with D shaped like forward() when D_has_hdim. Upstream uses ``repeat(self.D, "h -> h p", ...)``, which is only vali
tests/models/test_modeling_mamba2.py:205
Function_optional_restore_kernel
(x, y, N, BLOCK_SIZE: tl.constexpr)
tests/ops/test_cache.py:29
Method_post_hook
(kw, exception)
fla/ops/utils/cache.py:361
Method_pre_hook
(kw, reset_only=False)
fla/ops/utils/cache.py:353
Function_prepare_chunk_indices
(cu_seqlens, cu_seqlens_cpu=None)
tests/ops/utils/test_index.py:73
Method_prepare_input_fn
(sequence_sharding, mod, inputs, device_mesh)
fla/modules/layernorm.py:1442
Method_prepare_input_fn
( input_layouts, desired_input_layouts, mod, inputs, device_mesh, )
fla/modules/mlp.py:102
Method_prepare_input_fn
(sequence_sharding, mod, inputs, device_mesh)
fla/modules/fused_linear_cross_entropy.py:730
Method_prepare_output_fn
(use_local_output, mod, outputs, device_mesh)
fla/modules/layernorm.py:1462
Method_prepare_output_fn
(output_layouts, use_local_output, mod, outputs, device_mesh)
fla/modules/mlp.py:125
Method_prepare_output_fn
(use_local_output, mod, outputs, device_mesh)
fla/modules/fused_linear_cross_entropy.py:757
Method_replicate_module_fn
( self, name: str, module: nn.Module, device_mesh: DeviceMesh, )
fla/modules/layernorm.py:1430
Method_replicate_module_fn
( self, name: str, module: nn.Module, device_mesh: DeviceMesh, )
fla/modules/parallel.py:25
Function_rwkv7_post_init
RWKV7 needs a/b to be initialized as small positive values.
benchmarks/ops/registry.py:347
Function_silu_bwd_kernel
( y_ptr, dy_ptr, out_ptr, stride_y_n, stride_y_t, stride_y_d, stride_dy_n, str
fla/modules/backends/triton_ascend/causal_conv1d.py:226
Function_silu_kernel
( x_ptr, y_ptr, n_elements, BLOCK: tl.constexpr, )
fla/modules/backends/triton_ascend/causal_conv1d.py:177
Functionaddcmul_bwd_kernel1
( ixr, ixw, ixk, ixv, ixa, ixg, dxr, dxw, dxk, dxv, dxa, dxg,
fla/ops/rwkv7/fused_addcmul.py:115
Functionaddmm
( x: torch.Tensor, a: torch.Tensor, b: torch.Tensor, alpha: float | None = None, beta: flo
fla/ops/utils/matmul.py:214
Functionall_reduce_sum
All-reduce sum a tensor across ranks. Args: inp: Input tensor to reduce (modified in-place) group: Process group asy
fla/ops/cp/comm.py:44
Methodallocate_inference_cache
( self, batch_size: int, max_seqlen: int, device: torch.device | None = None,
fla/layers/mamba3.py:444
Functionargsort
( x, ids, dim: tl.constexpr = None, descending: tl.constexpr = tl.core.CONSTEXPR_0, )
fla/ops/nsa/utils.py:82
Functionattnres_bwd_kernel_dqdw
( q, w, dqw, dow_partial, dq, dw, dow, N, D: tl.constexpr, BN: tl.cons
fla/ops/attnres/fused.py:259
Functionattnres_bwd_kernel_dv
( q, res, w, ow, o_pre, rstd, logit, lse, do, dres, dqw, dow_p
fla/ops/attnres/fused.py:134
Functionattnres_fwd_kernel
( q, res, w, ow, o, o_pre, rstd, logit, lse, N, L, L2: tl.cons
fla/ops/attnres/fused.py:37
Functionautocast_to_fp16
(x)
fla/layers/rodimus.py:56
Functionbackward
(do, w, x)
fla/ops/deltaformer/invcum.py:33
Methodbackward
(ctx, dov, dht=None)
fla/ops/abc/chunk.py:953
Methodbackward
(ctx, d_output)
fla/ops/moba/parallel.py:223
Methodbackward
(ctx, do, dz_state)
fla/ops/linear_attn/utils.py:47
Methodbackward
(ctx, dy: torch.Tensor)
fla/ops/utils/pack.py:140
Methodbackward
(ctx, dy: torch.Tensor)
fla/ops/utils/pack.py:180
Methodbackward
( ctx, do, )
fla/ops/utils/pooling.py:187
Methodbackward
(ctx, do, dh_kk_final=None, dh_kv_final=None)
fla/ops/mesa_net/chunk.py:262
Methodbackward
(ctx, grad_output)
fla/ops/rwkv7/gate_output_correction.py:247
Methodbackward
(ctx, grad_output)
fla/ops/rwkv7/fused_k_update.py:340
Methodbackward
(ctx, dxr, dxw, dxk, dxv, dxa, dxg)
fla/ops/rwkv7/fused_addcmul.py:262
Methodbackward
(ctx, dkv)
fla/ops/rwkv7/channel_mixing.py:318
Methodbackward
( ctx, grad_u: torch.Tensor, )
fla/ops/deltaformer/parallel.py:700
Methodbackward
(ctx, do, dht=None)
fla/ops/hgrn/fused_recurrent.py:242
Methodbackward
(ctx, do, dht=None)
fla/ops/hgrn/chunk.py:252
Methodbackward
(ctx, do, dz)
fla/ops/based/parallel.py:350
Methodbackward
(ctx, do, dz)
fla/ops/based/fused_chunk.py:327
Methodbackward
(ctx, do)
fla/ops/wall_attn/parallel.py:872
Methodbackward
(ctx, do, dk_new)
fla/ops/path_attn/parallel.py:89
Methodbackward
(ctx, do)
fla/ops/attn/parallel.py:750
Methodbackward
(ctx, do: torch.Tensor, dht: torch.Tensor)
fla/ops/gdn2/chunk.py:133
Methodbackward
(ctx, do, dht)
fla/ops/generalized_delta_rule/iplr/fused_recurrent.py:338
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/generalized_delta_rule/iplr/chunk.py:418
Methodbackward
(ctx, do, dht)
fla/ops/generalized_delta_rule/dplr/fused_recurrent.py:193
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/generalized_delta_rule/dplr/chunk.py:235
Methodbackward
(ctx, do)
fla/ops/parallax/parallel.py:742
Methodbackward
(ctx, do, dht)
fla/ops/rwkv6/fused_recurrent.py:541
Methodbackward
(ctx, do, dht)
fla/ops/rwkv6/chunk.py:1223
Methodbackward
(ctx, do, dz)
fla/ops/rebased/parallel.py:397
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/gated_delta_product/chunk.py:194
Methodbackward
(ctx: FunctionCtx, gwkv: Tensor, gstate: Tensor)
fla/ops/rwkv4/fused_recurrent.py:469
Methodbackward
(ctx, dy: torch.Tensor)
fla/ops/common/gate.py:100
Methodbackward
(ctx, do, dht=None)
fla/ops/common/fused_chunk.py:569
Methodbackward
(ctx, do, dht)
fla/ops/common/fused_recurrent.py:599
Methodbackward
(ctx, do, dht)
fla/ops/gla/chunk.py:1317
Methodbackward
(ctx, dyg: torch.Tensor)
fla/ops/gated_delta_rule/gate.py:303
Methodbackward
(ctx, do, dht)
fla/ops/gated_delta_rule/fused_recurrent.py:299
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/gated_delta_rule/chunk.py:325
Methodbackward
(ctx, do, dht)
fla/ops/gated_oja_rule/fused_recurrent.py:211
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor )
fla/ops/gated_oja_rule/chunk.py:260
Methodbackward
( ctx, do: torch.Tensor, dp: torch.Tensor | None = None, )
fla/ops/attnres/fused.py:478
Methodbackward
(ctx, do, dht)
fla/ops/comba/fused_recurrent.py:215
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/comba/chunk.py:249
Methodbackward
(ctx, do)
fla/ops/nsa/parallel.py:762
Methodbackward
(ctx, do, *args)
fla/ops/nsa/compression.py:513
Methodbackward
(ctx, do, d_attn=None)
fla/ops/delta_rule/parallel.py:301
Methodbackward
(ctx, do, dht)
fla/ops/delta_rule/fused_recurrent.py:419
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/delta_rule/chunk.py:187
Methodbackward
(ctx, do, dht, dhbt)
fla/ops/ttt/fused_chunk.py:697
Methodbackward
(ctx, do, dht, dhbt)
fla/ops/ttt/chunk.py:1338
Methodbackward
(ctx, dyg: torch.Tensor)
fla/ops/kda/gate.py:308
Methodbackward
( ctx, do: torch.Tensor, dht: torch.Tensor, )
fla/ops/kda/chunk.py:124
Methodbackward
(ctx, do, dht)
fla/ops/log_linear_attn/chunk.py:1713
Methodbackward
(ctx, do, da=None)
fla/ops/simple_gla/parallel.py:654
Methodbackward
(ctx, do, dht)
fla/ops/simple_gla/chunk.py:206
Methodbackward
(ctx, do, dhkt=None, dhvt=None)
fla/ops/gsa/fused_recurrent.py:407
Methodbackward
(ctx, dov, dhkt=None, dhvt=None)
fla/ops/gsa/chunk.py:1032
Methodbackward
(ctx, dy, *args)
fla/modules/layernorm.py:786
Methodbackward
(ctx, dout, *args)
fla/modules/layernorm.py:1202
Methodbackward
(ctx, dout, *args)
fla/modules/fused_bitlinear.py:439
Methodbackward
(ctx, grad_losses, grad_z_losses)
fla/modules/fused_cross_entropy.py:294
← previousnext →1,301–1,400 of 2,989, ranked by callers