MCPcopy Create free account

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

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

↓ 1 callersFunctionchunk_generalized_iplr_delta_rule_fwd_o
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, v_new: torch.Tensor, b: torch.Tensor,
fla/ops/generalized_delta_rule/iplr/chunk.py:208
↓ 1 callersFunctionchunk_gla_bwd_dqk_intra
( q: torch.Tensor, k: torch.Tensor, g: torch.Tensor, dA: torch.Tensor, cu_seqlens: torch.L
fla/ops/gla/chunk.py:999
↓ 1 callersFunctionchunk_gla_bwd_dqkg
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, h: torch.Tensor, g: torch.Tensor,
fla/ops/gla/chunk.py:1042
↓ 1 callersFunctionchunk_gla_fwd_intra_gk
( q: torch.Tensor, k: torch.Tensor, g: torch.Tensor, scale: float, cu_seqlens: torch.LongT
fla/ops/gla/chunk.py:787
↓ 1 callersFunctionchunk_global_cumsum_scalar
( s: torch.Tensor, reverse: bool = False, cu_seqlens: torch.Tensor | None = None, scale: float
fla/ops/utils/cumsum.py:331
↓ 1 callersFunctionchunk_global_cumsum_vector
( s: torch.Tensor, reverse: bool = False, cu_seqlens: torch.Tensor | None = None, scale: float
fla/ops/utils/cumsum.py:362
↓ 1 callersFunctionchunk_gsa_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, s: torch.Tensor, g: torch.Tensor,
fla/ops/gsa/chunk.py:874
↓ 1 callersFunctionchunk_gsa_bwd_k
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, h: torch.Tensor,
fla/ops/gsa/chunk.py:666
↓ 1 callersFunctionchunk_gsa_bwd_v
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, h0: torch.Tensor,
fla/ops/gsa/chunk.py:631
↓ 1 callersFunctionchunk_gsa_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, s: torch.Tensor, g: torch.Tensor,
fla/ops/gsa/chunk.py:826
↓ 1 callersFunctionchunk_gsa_fwd_k
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, h0: torch.Tensor | N
fla/ops/gsa/chunk.py:547
↓ 1 callersFunctionchunk_gsa_fwd_v
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, scale: float = 1.,
fla/ops/gsa/chunk.py:519
↓ 1 callersFunctionchunk_iplr_delta_rule
r""" Args: q (torch.Tensor): queries of shape `[B, T, H, K]`. k (torch.Tensor): keys of shape `[B, T, H, K
fla/ops/generalized_delta_rule/iplr/chunk.py:430
↓ 1 callersFunctionchunk_kda_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, beta: torch.Tensor, Aqk: torch.Tensor
fla/ops/kda/chunk_bwd.py:419
↓ 1 callersFunctionchunk_kda_bwd_intra
( q: torch.Tensor, k: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, dAqk: torch.Tenso
fla/ops/kda/chunk_intra.py:866
↓ 1 callersFunctionchunk_kda_bwd_wy_dqkg_fused
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, v_new: torch.Tensor, g: torch.Tensor,
fla/ops/kda/chunk_bwd.py:352
↓ 1 callersFunctionchunk_kda_bwd_wy_dqkg_fused_tilelang
( q, k, v, v_new, g, beta, A, h, do, dh, dv, scale=None, cu_seqlens=None, chunk_size=64, chunk_ind
fla/ops/kda/backends/tilelang/chunk_bwd_dqkg.py:419
↓ 1 callersFunctionchunk_kda_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor,
fla/ops/kda/chunk_fwd.py:20
↓ 1 callersFunctionchunk_kda_fwd_intra
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, gk: torch.Tensor | None = None, beta:
fla/ops/kda/chunk_intra.py:764
↓ 1 callersFunctionchunk_kda_fwd_intra_token_parallel
Token-parallel implementation: each token gets its own thread block. Supports both fixed-length and variable-length sequences. Reduces wa
fla/ops/kda/chunk_intra_token_parallel.py:125
↓ 1 callersFunctionchunk_local_cumsum_scalar
( g: torch.Tensor, chunk_size: int, reverse: bool = False, scale: float = None, cu_seqlens
fla/ops/utils/cumsum.py:251
↓ 1 callersFunctionchunk_local_cumsum_vector
( g: torch.Tensor, chunk_size: int, reverse: bool = False, scale: float = None, cu_seqlens
fla/ops/utils/cumsum.py:288
↓ 1 callersFunctionchunk_mesa_cg_bwd
( dq: torch.Tensor, k: torch.Tensor, h: torch.Tensor, g_local_cumsum: torch.Tensor, beta:
fla/ops/mesa_net/chunk_cg_solver_bwd.py:122
↓ 1 callersFunctionchunk_mesa_cg_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, h: torch.Tensor, h_kv: torch.Tensor,
fla/ops/mesa_net/chunk_cg_solver_fwd.py:137
↓ 1 callersFunctionchunk_mesa_net_h_kk_bwd_intra_fn
( k: torch.Tensor, beta: torch.Tensor, g: torch.Tensor, h: torch.Tensor, dh: torch.Tensor,
fla/ops/mesa_net/chunk_h_kk_intra_bwd.py:140
↓ 1 callersFunctionchunk_mesa_net_h_kv_bwd_intra_fn
( q_star, k, v, beta, h_kv, dh_kv, g, do, cu_seqlens, chunk_size=64,
fla/ops/mesa_net/chunk_h_kv_intra_bwd.py:152
↓ 1 callersFunctionchunk_mesa_net_h_kv_bwd_intra_separate_fn
( q_star, k, v, beta, h_kv, dh_kv, g, do, cu_seqlens, chunk_size=64,
fla/ops/mesa_net/chunk_h_kv_intra_bwd_separate.py:236
↓ 1 callersFunctionchunk_oja_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, gv: torch.Tensor, beta: torch.Tensor,
fla/ops/gated_oja_rule/chunk.py:86
↓ 1 callersFunctionchunk_oja_bwd_dA
( v: torch.Tensor, gv: torch.Tensor, do: torch.Tensor, scale: float = 1., cu_seqlens: torc
fla/ops/gated_oja_rule/chunk_o.py:348
↓ 1 callersFunctionchunk_oja_bwd_dhu
( q: torch.Tensor, vg: torch.Tensor, w: torch.Tensor, do: torch.Tensor, dk: torch.Tensor,
fla/ops/gated_oja_rule/chunk_h.py:462
↓ 1 callersFunctionchunk_oja_bwd_dqk
( q: torch.Tensor, k: torch.Tensor, h: torch.Tensor, gv: torch.Tensor, dA: torch.Tensor,
fla/ops/gated_oja_rule/chunk_o.py:484
↓ 1 callersFunctionchunk_oja_bwd_dv_o
( v: torch.Tensor, gv: torch.Tensor, o: torch.Tensor, A: torch.Tensor, dv: torch.Tensor,
fla/ops/gated_oja_rule/chunk_o.py:634
↓ 1 callersFunctionchunk_oja_bwd_dvwg_h
( k: torch.Tensor, v: torch.Tensor, h: torch.Tensor, dh: torch.Tensor, dk: torch.Tensor,
fla/ops/gated_oja_rule/chunk_h.py:764
↓ 1 callersFunctionchunk_oja_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, gv: torch.Tensor, beta: torch.Tensor,
fla/ops/gated_oja_rule/chunk.py:25
↓ 1 callersFunctionchunk_oja_fwd_o
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, gv: torch.Tensor, h: torch.Tensor,
fla/ops/gated_oja_rule/chunk_o.py:184
↓ 1 callersFunctionchunk_oja_ref
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, beta: torch.Tensor,
tests/ops/test_oja.py:56
↓ 1 callersFunctionchunk_rwkv6_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, u: torch.Tensor,
fla/ops/rwkv6/chunk.py:1073
↓ 1 callersFunctionchunk_rwkv6_bwd_dh
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, gi: torch.Tensor, ge: torch.Tensor,
fla/ops/rwkv6/chunk.py:855
↓ 1 callersFunctionchunk_rwkv6_bwd_dqk_intra
( q: torch.Tensor, k: torch.Tensor, gi: torch.Tensor, ge: torch.Tensor, dA: torch.Tensor,
fla/ops/rwkv6/chunk.py:908
↓ 1 callersFunctionchunk_rwkv6_bwd_dqkgu
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, h: torch.Tensor, g: torch.Tensor,
fla/ops/rwkv6/chunk.py:953
↓ 1 callersFunctionchunk_rwkv6_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, u: torch.Tensor,
fla/ops/rwkv6/chunk.py:1014
↓ 1 callersFunctionchunk_rwkv6_fwd_intra
( q: torch.Tensor, k: torch.Tensor, gi: torch.Tensor, ge: torch.Tensor, u: torch.Tensor,
fla/ops/rwkv6/chunk.py:752
↓ 1 callersFunctionchunk_rwkv7
Args: r (torch.Tensor): r of shape `[B, T, H, K]`. w (torch.Tensor): log decay of shape `[B, T, H, K]`.
fla/ops/rwkv7/chunk.py:13
↓ 1 callersFunctionchunk_scaled_dot_comba_pkt_fwd
r""" Compute beta \mathcal{A}(i-1/j) * P * K^T. Args: k (torch.Tensor): The key tensor of shape `[B, T, H, K]`. p
fla/ops/comba/wy_fast.py:85
↓ 1 callersFunctionchunk_scaled_dot_kkt_bwd_gk
( k: torch.Tensor, g: torch.Tensor, beta: torch.Tensor, dA: torch.Tensor, cu_seqlens: torc
fla/ops/gated_oja_rule/chunk_kkt.py:480
↓ 1 callersFunctionchunk_scaled_dot_kkt_fwd
r""" Compute beta * K * K^T. Args: k (torch.Tensor): The key tensor of shape `[B, T, H, K]`. beta (torch.Tensor):
fla/ops/gated_oja_rule/chunk_kkt.py:386
↓ 1 callersFunctionchunk_simple_gla_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor, g_gamma: torch.Tenso
fla/ops/simple_gla/chunk.py:62
↓ 1 callersFunctionchunk_simple_gla_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, g: torch.Tensor | None = None, g_gamm
fla/ops/simple_gla/chunk.py:18
↓ 1 callersFunctionchunk_titans_linear
( q, k, v, w, b, theta, alpha, eta, eps, chunk_size, initial_state, output_final_state, )
fla/ops/titans/naive.py:216
↓ 1 callersFunctionchunk_transform_qk_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, beta: torch.Tensor, A: torch.Tensor,
fla/ops/delta_rule/parallel.py:85
↓ 1 callersFunctionchunk_ttt_linear_bwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, w: torch.Tensor, b: torch.Tensor,
fla/ops/ttt/chunk.py:1205
↓ 1 callersFunctionchunk_ttt_linear_bwd_dqke
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, eta: torch.Tensor, h: torch.Tensor,
fla/ops/ttt/chunk.py:1101
↓ 1 callersFunctionchunk_ttt_linear_bwd_dv_local
( q: torch.Tensor, k: torch.Tensor, eta: torch.Tensor, do: torch.Tensor, scale: float,
fla/ops/ttt/chunk.py:863
↓ 1 callersFunctionchunk_ttt_linear_bwd_h
( k: torch.Tensor, v: torch.Tensor, w: torch.Tensor, b: torch.Tensor, eta: torch.Tensor,
fla/ops/ttt/chunk.py:796
↓ 1 callersFunctionchunk_ttt_linear_bwd_norm
( q: torch.Tensor, # [B, H, L, D] k: torch.Tensor, # [B, H, L, D] v: torch.Tensor, # [B, H, L,
fla/ops/ttt/chunk.py:904
↓ 1 callersFunctionchunk_ttt_linear_fwd
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, w: torch.Tensor, b: torch.Tensor,
fla/ops/ttt/chunk.py:1159
↓ 1 callersFunctionchunk_ttt_linear_fwd_h
( k: torch.Tensor, v: torch.Tensor, w: torch.Tensor, b: torch.Tensor, eta: torch.Tensor,
fla/ops/ttt/chunk.py:677
↓ 1 callersFunctionchunk_ttt_linear_fwd_o
( q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, eta: torch.Tensor, h: torch.Tensor,
fla/ops/ttt/chunk.py:744
↓ 1 callersFunctionchunk_update_once
( b_p, b_k, b_v, b_m, b_g_exp_q, b_h, b_lamb, )
fla/ops/mesa_net/chunk_cg_solver_bwd.py:17
↓ 1 callersFunctionclear_triton_cache
Clear triton autotune cache to avoid cross-commit contamination.
scripts/run_benchmark_compare.py:169
↓ 1 callersFunctioncombine_params_log
log space Titians Parameters: - theta: θ - alpha: α - eta: η - seq_len: sequence length Returns: - beta, beta_T, f,
fla/ops/titans/log_impl.py:134
↓ 1 callersFunctioncompute_dh0_triton
Compute dh0 (gradient w.r.t. initial_state) using a separate Triton kernel. This is a workaround for Triton compiler bugs on some architectur
fla/modules/conv/triton/ops.py:100
↓ 1 callersFunctioncompute_subseq_len
Compute sub-sequence length for intracard splitting. For linear recurrence (fwd_h), the sequential scan is the bottleneck. Splitting always r
fla/ops/common/intracard_cp.py:139
↓ 1 callersFunctioncompute_x_k_grad
Args: dk1: (batch*seq_len, hidden_dim) x: (batch, seq_len, hidden_dim) x_prev: (batch, hidden_dim) or (batch, 1, hidden_d
fla/ops/rwkv7/channel_mixing.py:249
↓ 1 callersFunctionconstruct_H_matrix
(a, L)
fla/ops/log_linear_attn/naive.py:44
↓ 1 callersFunctionconstruct_level_mask
(level, L)
fla/ops/log_linear_attn/naive.py:21
↓ 1 callersFunctionconvert
( rwkv6: str, config: str, output: str, )
utils/convert_from_rwkv6.py:47
↓ 1 callersFunctionconvert
( rwkv7: str, output: str, precision: str = 'float32', )
utils/convert_from_rwkv7.py:21
↓ 1 callersFunctionconvert
( llama: str, config: str, output: str, precision: str = 'float32', )
utils/convert_from_llama.py:48
↓ 1 callersFunctioncopy_packages_to_output
Copy wheels and source distributions to output directory.
scripts/build_packages.py:259
↓ 1 callersFunctioncopy_runner_to_temp
Copy benchmark runner files to a temp directory for cross-commit use.
scripts/run_benchmark_compare.py:191
↓ 1 callersFunctioncross_entropy_loss
Arguments: logits: [batch, vocab_size] target: [batch,] label_smoothing: float logit_scale: float. Mu
fla/modules/fused_cross_entropy.py:327
↓ 1 callersFunctioncross_entropy_loss_npu
( logits: torch.Tensor, target: torch.Tensor, label_smoothing: float = 0.0, logit_scale: float
fla/modules/backends/triton_ascend/fused_cross_entropy.py:342
↓ 1 callersMethodcu2pad
(self, x, cu_seqlens)
fla/layers/mom.py:748
↓ 1 callersMethodcuda_kernels_forward
( self, hidden_states: torch.Tensor, last_state: dict | None = None, use_cache
fla/layers/log_linear_mamba2.py:439
↓ 1 callersMethodcuda_kernels_forward
( self, hidden_states: torch.Tensor, last_state: dict | None = None, use_cache
fla/layers/mamba.py:175
↓ 1 callersMethodcuda_kernels_forward
( self, hidden_states: torch.Tensor, last_state: dict | None = None, use_cache
fla/layers/mamba3.py:205
↓ 1 callersMethodcuda_kernels_forward
( self, hidden_states: torch.Tensor, last_state: dict | None = None, use_cache
fla/layers/mamba2.py:275
↓ 1 callersFunctioncumsum_comba_local_fwd_reference
(s, reverse=False, chunk_size=128)
tests/ops/test_comba.py:20
↓ 1 callersFunctioncustom_device_ctx
(index: int)
fla/utils/_device.py:200
↓ 1 callersFunctiondecorator
(fn: Callable[..., torch.Tensor])
fla/utils/_decorators.py:110
↓ 1 callersFunctiondplr_recurrence
(q, k, v, alpha, beta, gk, initial_state=None, output_final_state=True)
fla/ops/generalized_delta_rule/dplr/naive.py:16
↓ 1 callersMethodelu
(x)
fla/layers/linear_attn.py:94
↓ 1 callersMethodensure_initialized
Lazy-load backends on first use.
fla/ops/backends/__init__.py:139
↓ 1 callersMethodevaluate
Evaluate perplexity on the entire dataset
evals/ppl.py:118
↓ 1 callersMethodextra_repr
(self)
fla/modules/conv/short_conv.py:99
↓ 1 callersFunctionextract_configs
Extract all autotune configs from Triton cache.
scripts/utils/autotune_export.py:241
↓ 1 callersFunctionextract_default_config
(config: object)
scripts/utils/autotune_export.py:74
↓ 1 callersFunctionextract_dependencies
Read base dependencies and optional-dependency extras from pyproject.toml.
scripts/build_packages.py:22
↓ 1 callersFunctionextract_gpu_memory_used
Extract 'GPU Memory Used (MiB)' from xpu-smi output.
scripts/check_gpu.py:29
↓ 1 callersFunctionfile_to_module_path
Convert file path to Python module path.
scripts/find_dependent_tests.py:130
↓ 1 callersMethodfilter
(self, seq_len: int, *args, **kwargs)
fla/modules/conv/long_conv.py:156
↓ 1 callersFunctionfind_affected_op_names
(changed_files: list[str])
scripts/run_benchmark_compare.py:107
↓ 1 callersMethodfind_changed_entries
Return entries from existing whose key+config differs from self (for backup).
scripts/utils/autotune_export.py:176
↓ 1 callersMethodfind_dependent_tests
(self, changed_files_str: list, max_depth=4)
scripts/find_dependent_tests.py:204
↓ 1 callersFunctionfind_dispatch_op_files
Find all files in fla/ that define a function with @dispatch and name in methods.
scripts/find_dependent_tests.py:102
↓ 1 callersFunctionformat_kernel_table
Format kernel stats as a table.
benchmarks/cp/benchmark_kda_cp8_vs_cp2tp.py:141
↓ 1 callersFunctionforward
(u, w)
fla/ops/deltaformer/invcum.py:11
↓ 1 callersMethodfrom_env
(cls)
fla/ops/utils/cache.py:51
↓ 1 callersMethodfrom_file
(cls, autotune_file: Path)
scripts/utils/autotune_export.py:40
← previousnext →601–700 of 2,989, ranked by callers