Methodbwd_inner(q, z, do, B, H, T, K, V, BT, BK, BV, NT, scale, normk=False)
fla/ops/abc/chunk.py:965
Methodbwd_post(s, z, ss, B, H, T, S, BT, BC, BS, NT, NC, NS)
fla/ops/abc/chunk.py:979
Methodcausal_conv1d_bwd(
self,
x,
dy,
dht,
weight=None,
bias=None,
residual=N
fla/modules/backends/triton_ascend/__init__.py:308
Functioncausal_conv1d_bwd_kernel(
x,
weight,
initial_state,
dht,
dy,
dx,
dw,
db,
cu_seqlens,
chunk_ind
fla/modules/backends/triton_ascend/causal_conv1d.py:396
Functioncausal_conv1d_bwd_kernel(
x,
y,
weight,
initial_state,
dht,
dy,
dx,
dw,
db,
cu_seqlens,
ch
fla/modules/conv/triton/kernels.py:154
Functioncausal_conv1d_bwd_seq_kernel(
x,
weight,
dy,
dx,
dw,
db,
stride_x_n,
stride_x_t,
stride_x_d,
strid
fla/modules/backends/triton_ascend/causal_conv1d.py:318
Methodcausal_conv1d_fwd(
self,
x,
weight,
bias,
residual,
initial_state=None,
fla/modules/backends/triton_ascend/__init__.py:277
Functioncausal_conv1d_fwd_kernel(
x,
y,
weight,
bias,
cu_seqlens,
initial_state,
chunk_indices,
B,
T,
fla/modules/backends/triton_ascend/causal_conv1d.py:93
Functioncausal_conv1d_fwd_kernel(
x,
y,
weight,
bias,
residual,
cu_seqlens,
initial_state,
chunk_indices,
fla/modules/conv/triton/kernels.py:37
Functioncausal_conv1d_states_fwd_kernel(
x,
initial_state,
final_state,
cu_seqlens,
T,
D,
W,
stride_x_n,
stride_x
fla/modules/backends/triton_ascend/causal_conv1d.py:592
Functioncausal_conv1d_states_fwd_kernel(
x,
initial_state,
final_state,
cu_seqlens,
T,
D,
W,
stride_x_n,
stride_x
fla/modules/conv/triton/kernels.py:510
Functioncausal_conv1d_update_kernel(
x,
cache,
y,
weight,
bias,
stride_x_n,
stride_x_d,
stride_y_n,
stride_y_
fla/modules/backends/triton_ascend/causal_conv1d.py:653
Functioncausal_conv1d_update_kernel(
x,
cache,
residual,
y,
weight,
bias,
stride_x_n, # batch stride
stride_x_d,
fla/modules/conv/triton/kernels.py:348
Methodcausal_conv1d_update_states(
self,
x,
state_len,
initial_state=None,
cu_seqlens=None,
)
fla/modules/backends/triton_ascend/__init__.py:360
Functionchunk_abc_bwd_kernel_K(
q,
k,
v,
z,
h,
A,
do,
dh,
dq,
dk,
dv,
dA,
scale,
T,
fla/ops/abc/chunk.py:605
Functionchunk_abc_bwd_kernel_V(
k,
v,
z,
h,
A,
do,
dh,
dq,
dk,
dv,
dA,
scale,
T,
K:
fla/ops/abc/chunk.py:360
Functionchunk_abc_bwd_kernel_dh(
q,
z,
do,
dh,
scale,
T,
K: tl.constexpr,
V: tl.constexpr,
BT: tl.constex
fla/ops/abc/chunk.py:301
Functionchunk_abc_bwd_kernel_intra_K(
v,
z,
do,
dA,
scale,
T,
V: tl.constexpr,
BT: tl.constexpr,
BC: tl.conste
fla/ops/abc/chunk.py:547
Functionchunk_abc_bwd_kernel_intra_KV(
v,
z,
A,
do,
dv,
T,
V: tl.constexpr,
BT: tl.constexpr,
BC: tl.constexpr,
fla/ops/abc/chunk.py:695
Functionchunk_abc_bwd_kernel_intra_V(
q,
k,
z,
dA,
dq,
dk,
T,
K: tl.constexpr,
BT: tl.constexpr,
BC: tl.co
fla/ops/abc/chunk.py:454
Functionchunk_abc_bwd_kernel_rcum_inter(
s,
z,
ss,
doo,
T,
S: tl.constexpr,
BT: tl.constexpr,
BS: tl.constexpr,
N
fla/ops/abc/chunk.py:749
Functionchunk_abc_bwd_kernel_rcum_intra(
s,
z,
ss,
doo,
T,
S: tl.constexpr,
BT: tl.constexpr,
BC: tl.constexpr,
B
fla/ops/abc/chunk.py:785
Functionchunk_abc_fwd_kernel_K(
q,
k,
z,
h,
o,
A,
scale,
T,
K: tl.constexpr,
V: tl.constexpr,
BT
fla/ops/abc/chunk.py:133
Functionchunk_abc_fwd_kernel_V(
q,
v,
z,
h,
o,
A,
scale,
T,
K: tl.constexpr,
V: tl.constexpr,
BT
fla/ops/abc/chunk.py:249
Functionchunk_abc_fwd_kernel_h(
k,
v,
z,
h,
h0,
ht,
T,
K: tl.constexpr,
V: tl.constexpr,
BT: tl.cons
fla/ops/abc/chunk.py:19
Functionchunk_abc_fwd_kernel_intra_K(
v,
z,
o,
A,
T,
V: tl.constexpr,
BT: tl.constexpr,
BC: tl.constexpr,
BV:
fla/ops/abc/chunk.py:83
Functionchunk_abc_fwd_kernel_intra_V(
q,
k,
z,
A,
scale,
T,
K: tl.constexpr,
BT: tl.constexpr,
BC: tl.constexp
fla/ops/abc/chunk.py:191
Functionchunk_bwd_dh(
q: torch.Tensor,
k: torch.Tensor,
v: torch.Tensor,
g: torch.Tensor,
gk: torch.Tensor,
fla/ops/common/chunk_h_parallel.py:482
Functionchunk_bwd_dh(
q: torch.Tensor,
k: torch.Tensor,
v: torch.Tensor,
g: torch.Tensor,
gk: torch.Tensor,
fla/ops/common/chunk_h_split.py:512
Functionchunk_bwd_kernel_dh(
q,
g,
g_gamma,
gk,
gv,
do,
dh,
dht,
dh0,
cu_seqlens,
split_offse
fla/ops/common/chunk_h.py:173
Functionchunk_bwd_kernel_dh_parallel(
q,
g,
gk,
gv,
do,
dh,
dht,
dh0,
cu_seqlens,
chunk_indices,
scale
fla/ops/common/chunk_h_parallel.py:240
Functionchunk_bwd_kernel_dh_reduction(
g,
gk,
gv,
dh,
doq0,
dh0,
cu_seqlens,
chunk_offsets,
T,
HQ: tl.const
fla/ops/common/chunk_h_parallel.py:344
Functionchunk_bwd_kernel_dh_reduction(
g,
gk,
gv,
dhs,
dhr,
dh0,
cu_seqlens,
split_offsets,
T,
S: tl.conste
fla/ops/common/chunk_h_split.py:359
Functionchunk_bwd_kernel_dh_split(
q,
g,
gk,
gv,
do,
dht,
dhs,
dhr,
dh0,
cu_seqlens,
split_indices,
fla/ops/common/chunk_h_split.py:240
Functionchunk_bwd_kernel_dqke(
q,
k,
v,
e,
h,
do,
dh,
dhb,
dq,
dk,
de,
cu_seqlens,
chun
fla/ops/ttt/chunk.py:566
Functionchunk_bwd_kernel_dqkwg(
q,
k,
v,
g,
g_gamma,
h,
do,
dh,
dq,
dk,
dw,
dv,
dg,
fla/ops/common/chunk_o.py:148
Functionchunk_bwd_kernel_dv(
q,
k,
g,
g_gamma,
do,
dv,
dh,
cu_seqlens,
chunk_indices,
scale,
fla/ops/common/chunk_o.py:329
Functionchunk_bwd_kernel_dv_local(
q,
k,
g,
g_gamma,
A,
do,
dv,
cu_seqlens,
chunk_indices,
scale,
T
fla/ops/common/chunk_o.py:431
Functionchunk_comba_cumsum_scalar_bwd_kernel(
dg0,
dgr,
cu_seqlens,
chunk_indices,
T,
B: tl.constexpr,
H: tl.constexpr,
BT
fla/ops/comba/utils.py:108
Functionchunk_comba_cumsum_scalar_fwd_kernel(
g,
g0,
g1,
scale,
cu_seqlens,
chunk_indices,
T,
B: tl.constexpr,
H: tl.c
fla/ops/comba/utils.py:29
Functionchunk_cumprod_householder_bwd_kernel(
hc_suffix, dhc_whole,
k, dk, w1, w2, dw1, dw2, dk_new,
cu_seqlens, split_indices, chunk_offsets,
fla/ops/path_attn/cumprod_householder_bwd.py:20
Functionchunk_cumprod_householder_fwd_kernel(
k,
k_new,
w1,
w2,
hc_suffix,
hc_whole,
cu_seqlens,
split_indices,
chunk_
fla/ops/path_attn/cumprod_householder_fwd.py:20
Functionchunk_dplr_bwd_dgk_kernel(
dgk,
dgk_offset,
dgk_last,
dgk_output,
cu_seqlens,
chunk_indices,
T,
H: tl.c
fla/ops/generalized_delta_rule/dplr/chunk_A_bwd.py:408
Functionchunk_dplr_bwd_kernel_dAu(
v,
do,
v_new,
A_qb,
dA_qk,
dA_qb,
dv_new,
cu_seqlens,
chunk_indices,
fla/ops/generalized_delta_rule/dplr/chunk_o_bwd.py:35
Functionchunk_dplr_bwd_kernel_dhu(
qg,
bg,
w,
gk,
dht,
dh0,
do,
dh,
dv,
dv2,
cu_seqlens,
chunk_
fla/ops/generalized_delta_rule/dplr/chunk_h_bwd.py:35
Functionchunk_dplr_bwd_kernel_dv(
A_qk,
kg,
do,
dv,
dh,
cu_seqlens,
chunk_indices,
T,
H: tl.constexpr,
fla/ops/generalized_delta_rule/dplr/chunk_o_bwd.py:238
Functionchunk_dplr_bwd_kernel_intra(
q,
k,
a,
b,
gi,
ge,
dAqk,
dAqb,
dAak,
dAab,
dq,
dk,
da,
fla/ops/generalized_delta_rule/dplr/chunk_A_bwd.py:33
Functionchunk_dplr_bwd_kernel_intra_tensorcore(
q,
k,
a,
b,
gi,
ge,
dAqk,
dAqb,
dAak,
dAab,
dq,
dk,
da,
fla/ops/generalized_delta_rule/dplr/chunk_A_bwd.py:238
Functionchunk_dplr_bwd_o_kernel(
v,
v_new,
h,
do,
dh,
dk,
db,
w,
dq,
dv,
dw,
gk,
dgk_last
fla/ops/generalized_delta_rule/dplr/chunk_o_bwd.py:108
Functionchunk_dplr_fwd_A_kernel_intra_sub_intra(
q,
k,
a,
b,
gi,
ge,
qg,
kg,
ag,
bg,
Aqk,
Aqb,
Aab,
A
fla/ops/generalized_delta_rule/dplr/chunk_A_fwd.py:33
Functionchunk_dplr_fwd_A_kernel_intra_tensorcore(
q,
k,
a,
b,
gi,
ge,
qg,
kg,
ag,
bg,
Aqk,
Aqb,
Aab,
A
fla/ops/generalized_delta_rule/dplr/chunk_A_fwd.py:160
Functionchunk_dplr_fwd_kernel_h(
kg,
v,
w,
bg,
u,
v_new,
gk,
h,
h0,
ht,
cu_seqlens,
chunk_off
fla/ops/generalized_delta_rule/dplr/chunk_h_fwd.py:35
Functionchunk_dplr_fwd_kernel_o(
qg,
v,
v_new,
A_qk,
A_qb,
h,
o,
cu_seqlens,
chunk_indices,
T,
H:
fla/ops/generalized_delta_rule/dplr/chunk_o_fwd.py:36
Functionchunk_fwd_h(
k: torch.Tensor,
v: torch.Tensor,
g: torch.Tensor,
gk: torch.Tensor,
gv: torch.Tensor,
fla/ops/common/chunk_h_parallel.py:411