MCPcopy Create free account

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

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

Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_abc.py:67
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_forgetting_transformer.py:67
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_hgrn2.py:67
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_lightnet.py:67
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_path_attn.py:68
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, self_decoder_attn_type: str, dtype: torc
tests/models/test_modeling_yoco.py:143
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_mamba.py:60
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_gated_deltanet.py:66
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_rwkv6.py:54
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_moba.py:80
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_mesanet.py:67
Functiontest_generation
( L: int, B: int, T: int, use_forget_gate: bool, num_householders: int, dtype: torch.d
tests/models/test_modeling_gated_deltaproduct.py:71
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_hgrn.py:67
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, conv_backend: str, )
tests/models/test_modeling_mamba2.py:103
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_comba.py:67
Functiontest_generation
( L: int, B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/models/test_modeling_transformer.py:68
Functiontest_global_cumsum
( B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/ops/utils/test_cumsum.py:38
Functiontest_global_cumsum_varlen
( H: int, D: int, cu_seqlens: list[int], dtype: torch.dtype, )
tests/ops/utils/test_cumsum.py:74
Functiontest_global_reversed_cumsum
( B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/ops/utils/test_cumsum.py:114
Functiontest_global_reversed_cumsum_varlen
( H: int, D: int, cu_seqlens: list[int], dtype: torch.dtype, )
tests/ops/utils/test_cumsum.py:150
Functiontest_groupnorm
(B: int, T: int, D: int, G: int, is_rms_norm: bool)
tests/modules/test_layernorm.py:61
Functiontest_groupnorm_linear
(N: int, D: int, G: int, is_rms_norm: bool)
tests/modules/test_layernorm.py:155
Functiontest_groupnorm_small_t
GroupNorm backward must handle T < SM_count without illegal memory access.
tests/modules/test_layernorm.py:301
Functiontest_hgrn2_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_hgrn2.py:32
Functiontest_inference
( B: int, T: int, HQ: int, H: int, D: int, M: int, dtype: torch.dtype, )
tests/ops/test_gsa.py:405
Functiontest_input_guard_makes_tensors_contiguous_except_skipped
()
tests/test_utils.py:49
Functiontest_intracard_backend_disabled_by_default
Verify that IntraCardCPBackend is disabled by default.
tests/ops/test_intracard_cache.py:105
Functiontest_intracard_backend_disabled_when_env_var_is_zero
Verify that IntraCardCPBackend is disabled when FLA_INTRACARD_CP=0.
tests/ops/test_intracard_cache.py:113
Functiontest_intracard_backend_enabled_when_env_var_is_one
Verify that IntraCardCPBackend is enabled when FLA_INTRACARD_CP=1.
tests/ops/test_intracard_cache.py:121
Functiontest_is_inner_contiguous
Test _is_inner_contiguous correctly classifies tensor layouts.
tests/modules/test_activation.py:359
Functiontest_l2norm
(B: int, T: int, H: int, D: int, dtype: torch.dtype)
tests/modules/test_l2norm.py:32
Functiontest_layer
The GatedDeltaNet2 layer must run fwd/bwd and produce finite grads for every parameter, including the GVA (num_v_heads > num_heads) and short-conv
tests/ops/test_gdn2.py:440
Functiontest_layernorm
(B: int, H: int, T: int, D: int, elementwise_affine: bool, bias: bool)
tests/modules/test_layernorm.py:25
Functiontest_layernorm_gated
(B: int, H: int, T: int, D: int, elementwise_affine: bool, activation: str, bias: bool)
tests/modules/test_layernorm_gated.py:29
Functiontest_layernorm_linear
(N: int, D: int)
tests/modules/test_layernorm.py:117
Functiontest_layernorm_small_t
LayerNorm backward must handle T < SM_count without illegal memory access.
tests/modules/test_layernorm.py:270
Functiontest_layers_without_cache_still_work_with_layer_idx_none
(builder, hidden_states)
tests/layers/test_layer_cache_layer_idx.py:218
Functiontest_legacy_cache_decode_all_layers_see_same_past_length
Test that LegacyFLACache also has consistent past length across all layers during decode. Same issue as test_cache_decode_all_layers_see_sam
tests/models/test_cache.py:426
Functiontest_legacy_cache_window_overflow_without_hf_init
()
tests/models/test_cache.py:234
Functiontest_local_cumsum
( B: int, T: int, H: int, C: int, D: int, dtype: torch.dtype, )
tests/ops/utils/test_cumsum.py:186
Functiontest_local_cumsum_varlen
( H: int, C: int, D: int, cu_seqlens: list[int], dtype: torch.dtype, )
tests/ops/utils/test_cumsum.py:223
Functiontest_logsigmoid
(B: int, T: int, D: int, temperature: float, noncontiguous: bool, compile: bool)
tests/modules/test_activation.py:84
Functiontest_mamba2_layer_gradient_vs_official
Gradients of custom Mamba2 (train) must match official mamba_ssm.Mamba2. Both use the fused mamba_split_conv1d_scan_combined kernel in train mode
tests/models/test_modeling_mamba2.py:482
Functiontest_mamba2_layer_vs_official_eval
Eval-mode output of custom Mamba2 must match official mamba_ssm.Mamba2. Both use the non-fused path: causal_conv1d_fn + mamba_chunk_scan_combined
tests/models/test_modeling_mamba2.py:427
Functiontest_mamba2_layer_vs_official_inference
Step-by-step decode: FLA cached forward vs ``mamba_ssm.Mamba2.step`` with rolling ``(conv_state, ssm_state)``. When ``D_has_hdim`` is True, t
tests/models/test_modeling_mamba2.py:307
Functiontest_mamba2_layer_vs_official_train
Training-mode output of custom Mamba2 must match official mamba_ssm.Mamba2. Both should use mamba_split_conv1d_scan_combined (the fused kernel) i
tests/models/test_modeling_mamba2.py:378
Functiontest_mamba_layer_gradient_vs_official
Gradients of custom Mamba (train) must match official mamba_ssm.Mamba. Both use mamba_inner_fn in train mode, so gradients flow through the
tests/models/test_modeling_mamba.py:258
Functiontest_mamba_layer_vs_official_eval
Eval-mode output of custom Mamba must match official mamba_ssm.Mamba. Custom uses selective_scan_fn in eval; official uses mamba_inner_fn.
tests/models/test_modeling_mamba.py:223
Functiontest_mamba_layer_vs_official_inference
Step-by-step inference comparison between custom Mamba and official mamba_ssm.Mamba. Starting with empty inference params and passing in iden
tests/models/test_modeling_mamba.py:137
Functiontest_mamba_layer_vs_official_train
Training-mode output of custom Mamba must match official mamba_ssm.Mamba. Both should use mamba_inner_fn (the fused CUDA kernel) in this mod
tests/models/test_modeling_mamba.py:189
Functiontest_matches_reference
(name, impl, ref, batch_size, max_seq_len)
tests/ops/utils/test_index.py:120
Functiontest_mean_pooling
( B: int, T: int, H: int, C: int, D: int, dtype: torch.dtype, )
tests/ops/utils/test_pooling.py:30
Functiontest_mean_pooling_varlen
( H: int, C: int, D: int, cu_seqlens: list[int], dtype: torch.dtype, )
tests/ops/utils/test_pooling.py:71
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_mom.py:33
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_gsa.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_samba.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_deltanet.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_linear_attn.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_retnet.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_parallax.py:33
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_gla.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, dtype: torch.dtype, )
tests/models/test_modeling_rodimus.py:30
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_mla.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_nsa.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_bitnet.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, dtype: torch.dtype, )
tests/models/test_modeling_rwkv7.py:30
Functiontest_modeling
Test the forward and backward pass of the Mamba3 model by manually instantiating the configuration and the model.
tests/models/test_modeling_mamba3.py:50
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, expand_v: float, use_l2warp: bool, d
tests/models/test_modeling_raven.py:28
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, use_scalar_gate: bool,
tests/models/test_modeling_wall_transformer.py:31
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_deltaformer.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_kda.py:31
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_abc.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_forgetting_transformer.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_lightnet.py:32
Functiontest_modeling
Test the forward and backward pass of the Mamba2 model by manually instantiating the configuration and the model.
tests/models/test_modeling_log_linear_mamba2.py:33
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_path_attn.py:31
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, self_decoder_attn_type: str, use_l2warp:
tests/models/test_modeling_yoco.py:98
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, dtype: torch.dtype, )
tests/models/test_modeling_mamba.py:36
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_gated_deltanet.py:31
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, dtype: torch.dtype, )
tests/models/test_modeling_rwkv6.py:30
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, qk_norm: bool, use
tests/models/test_modeling_moba.py:38
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_mesanet.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_gated_deltaproduct.py:35
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_hgrn.py:32
Functiontest_modeling
Test the forward and backward pass of the Mamba2 model by manually instantiating the configuration and the model.
tests/models/test_modeling_mamba2.py:41
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_comba.py:32
Functiontest_modeling
( L: int, B: int, T: int, H: int, D: int, use_l2warp: bool, attnres_block_size: in
tests/models/test_modeling_transformer.py:33
Functiontest_naive_chunk
( B: int, T: int, H: int, D: int, dtype: torch.dtype, )
tests/ops/test_titans.py:61
Functiontest_naive_chunk
( B: int, T: int, H: int, D: int, normalize: bool, dtype: torch.dtype, )
tests/ops/test_linear_attn.py:81
Functiontest_naive_chunk
( B: int, T: int, H: int, HV: int, D: int, scale: float, gate_logit_normalizer: fl
tests/ops/test_kda.py:40
Functiontest_naive_chunk_oja
( B: int, T: int, H: int, D: int, scale: float, gate_logit_normalizer: float, dtyp
tests/ops/test_oja.py:155
Functiontest_no_recompile_across_multi_segment_shapes
(name, impl, ref)
tests/ops/utils/test_index.py:173
Functiontest_no_recompile_across_single_segment_shapes
(name, impl, ref)
tests/ops/utils/test_index.py:180
Functiontest_normalize_grad
Gradient parity for `normalize=True` against the naive recurrent reference. Regression test for the missing `dk` denominator-path contribution: p
tests/ops/test_linear_attn.py:273
Functiontest_normalize_split_resume
Splitting at `split` and resuming via the returned (kv_state, z_state) must reproduce the single-call output when normalize=True.
tests/ops/test_linear_attn.py:239
Functiontest_normalize_zinit_grad
Gradient parity when chaining via `(kv_state, z_state)`: splitting at `split` and resuming with the prior `z_state` as `z_init` must produce the
tests/ops/test_linear_attn.py:318
Functiontest_pack_sequence
( B: int, T: int, H: int, D: int, padding_side: str, dtype: torch.dtype, )
tests/ops/utils/test_pack.py:29
Functiontest_parallel
( B: int, T: int, H: int, HQ: int, D: int, scale: float, )
tests/ops/test_attn.py:31
Functiontest_parallel
( B: int, H: int, HQ: int, T: int, D: int, use_forget_gate: bool, dtype: torch.dty
tests/ops/test_path_attn.py:37
Functiontest_parallel
( B: int, T: int, H: int, HQ: int, D: int, S: int, block_size: int, scale: flo
tests/ops/test_nsa.py:34
← previousnext →2,801–2,900 of 2,989, ranked by callers