Functions — github.com/NVlabs/LongLive

CodeHub MCPcopy Create free account

hub / github.com/NVlabs/LongLive / functions

Functions963 in github.com/NVlabs/LongLive

⨍ Functions 963◇ Types & classes 175↳ Endpoints 2

( self, num_buckets=40, max_size_per_bucket=50, num_train_timesteps=1000,

utils/error_buffer.py:41

(self, num_inference_steps=100, num_train_timesteps=1000, shift=3.0, sigma_max=1.0, sigma_min=0.003 / 1.002, i

utils/scheduler.py:108

(self, fsdp_module: torch.nn.Module, decay: float = 0.999)

utils/distributed.py:96

(self, prompt_path, extended_prompt_path=None)

utils/dataset.py:26

(self, prompt_path: str, field: str = "prompts", cache_dir: str | None = None)

utils/dataset.py:66

( self, data_path: str, num_blocks: int, chunks_per_shot: int = 0, sce

utils/dataset.py:125

( self, data_dir, video_size, total_frames, target_fps=16, vid

utils/dataset.py:280

( self, fn, *, name: str, backend: str = "inductor", mode: str

utils/torch_compile_utils.py:26

utils/wan_5b_wrapper.py:17

( self, model_name="Wan2.2-TI2V-5B", timestep_shift=8.0, is_ca

utils/wan_5b_wrapper.py:278

( self, vae_path: str, pruning_rate: Optional[float] = None, dtype: torch.dtyp

utils/lightvae_5b_wrapper.py:342

(self, args, device)

model/base.py:145

Initialize the Diffusion loss module.

model/diffusion.py:15

Initialize the DMD (Distribution Matching Distillation) module. This class is self-contained and compute generator and fake score los

model/dmd.py:20

(self, *args: list[Any], **kwargs: dict[str, Any])

fouroversix/setup.py:177

( self, module: GptOssExperts, quantization_config: ModuleQuantizationConfig | None =

fouroversix/src/fouroversix/model/modules/gpt_oss.py:93

Initialize the FourOverSixLinear layer. Args: module (nn.Linear): The high-precision module that this quantized layer wil

fouroversix/src/fouroversix/model/modules/linear.py:159

Initialize the FourOverSixLinear layer. Args: module (nn.Linear): The high-precision module that this quantized layer wil

fouroversix/src/fouroversix/model/modules/linear_ori.py:123

Initialize the FourOverSixQwenExperts layer. Args: module (GptOssMLP): The high-precision module that this quantized lay

fouroversix/src/fouroversix/model/modules/qwen.py:30

( self, values: torch.Tensor, scale_factors: torch.Tensor, amax: torch.Tensor,

fouroversix/src/fouroversix/quantize/quantized_tensor.py:70

( self, dtype: DataType = None, scale_rule: ScaleRule = None, )

fouroversix/src/fouroversix/weight_conversions/gpt_oss.py:12

(self, group_name: str | None = None)

fouroversix/scripts/ptq/coordinators/local.py:18

( self, *args: list[Any], smoothquant_alpha: float, **kwargs: dict[str, Any],

fouroversix/scripts/ptq/evaluators/smoothquant.py:40

( # noqa: C901 self, model_name: str, model: AutoModelForCausalLM, config: Ge

fouroversix/scripts/ptq/evaluators/utils.py:30

(self, *args: list[Any], **kwargs: dict[str, Any])

fouroversix/scripts/ptq/evaluators/awq.py:34

tests/test_i2v_teacher_forcing_context.py:28

trainer/distillation.py:37

trainer/diffusion.py:56

( self, args, device, generator=None, text_encoder

pipeline/causal_diffusion_inference.py:30

(self, scheduler: SchedulerInterface, generator: WanDiffusionWrapper,

pipeline/self_forcing_training.py:17

( self, args, device, generator=None, text_encoder=None, vae=N

pipeline/causal_diffusion_inference_sp.py:167

wan_5b/utils/fm_solvers.py:853

wan_5b/utils/fm_solvers_unipc.py:801

utils/dataset.py:37

utils/dataset.py:94

utils/dataset.py:154

utils/dataset.py:743

Method__post_init__

utils/quant.py:78

Method__post_init__

Convert string values to enums.

fouroversix/src/fouroversix/model/config.py:62

Method__post_init__

Convert module config overrides to ModuleQuantizationConfig instances.

fouroversix/src/fouroversix/model/config.py:170

Method__post_init__

Convert string values to enums.

fouroversix/src/fouroversix/quantize/config.py:38

Function_adaln_modulate_kernel

( x_ptr, # [B, L, C] scale_ptr, # [B, F, 1, C] (or any layout, indexed via strides) shif

utils/adaln_triton.py:38

Method_chunk_and_attend

wan_5b/modules/causal_model.py:275

Method_cross_attend

wan_5b/modules/causal_model.py:223

Method_dit_model

Return the underlying CausalWanModel, unwrapping PeftModel if present. After LoRA wrapping, ``self.generator.model`` is a PeftModel whose

pipeline/causal_diffusion_inference.py:153

Method_get_timestep

( self, min_timestep, max_timestep, batch_size, num_frame, num

tests/test_i2v_teacher_forcing_context.py:67

wan_5b/modules/causal_model_sp_ulysses.py:592

Method_initialize_inference_pipeline

Lazy initialize the inference pipeline during the first backward simulation run. Here we encapsulate the inference code with a model-

model/diffusion.py:562

Method_initialize_kv_cache

Initialize head-split KV caches for Ulysses SP.

pipeline/causal_diffusion_inference_sp.py:206

Method_initialize_models

(self, args, device)

model/diffusion.py:122

Function_materialize_mixed_quantized_weights_for_inference

( model: nn.Module, target_device: torch.device | str | None = None, cache_transposed_weights: boo

utils/quant.py:450

Method_move_optimizer_to_device

Move optimizer state to the specified device.

trainer/distillation.py:790

Method_move_optimizer_to_device

Move optimizer state to the specified device.

trainer/diffusion.py:545

Function_read_video_decord

read video using decord.VideoReader Args: ele (dict): a dict contains the configuration of video. support keys: - vid

wan_5b/utils/qwen_vl_utils.py:221

Function_read_video_torchvision

read video using torchvision.io.read_video Args: ele (dict): a dict contains the configuration of video. support keys:

wan_5b/utils/qwen_vl_utils.py:177

Method_resolve_buf_file

trainer/diffusion.py:518

Function_rope_apply_kernel

( x_ptr, # [S_total, N, D] bf16 (D is even, pairs are (a,b)=(2d, 2d+1)) cos_ptr, # [seq

utils/rope_triton.py:37

Method_set_gradient_checkpointing

(self, module, value=False)

wan_5b/modules/causal_model.py:1100

Method_set_gradient_checkpointing

(self, module, value=False)

wan_5b/modules/model.py:433

Function_shim_read_video

(filename, pts_unit="sec", output_format="THWC", **_unused)

inference.py:26

Function_shim_write_video

(filename, video_array, fps, **_unused)

inference.py:16

Method_sigma_to_t

wan_5b/utils/fm_solvers.py:327

Method_sigma_to_t

wan_5b/utils/fm_solvers_unipc.py:271

Method_update_sink_for_scene_cut

Legacy copy-to-front sink relocation (used by training pipeline).

pipeline/causal_diffusion_inference.py:1001

Method_vae_thread_fn

pipeline/causal_diffusion_inference.py:468

Methodadd_custom_field

(self, key: str, value)

wan_5b/utils/prompt_extend.py:60

Methodadd_noise

( self, original_samples: torch.Tensor, noise: torch.Tensor, timesteps: torch.

wan_5b/utils/fm_solvers.py:812

Methodadd_noise

( self, original_samples: torch.Tensor, noise: torch.Tensor, timesteps: torch.

wan_5b/utils/fm_solvers_unipc.py:760

Methodadd_noise

Diffusion forward corruption process. Input: - clean_latent: the clean latent with shape [B*T, C, H, W] - noi

utils/scheduler.py:159

Methodadd_noise

(self, clean, noise, timestep)

tests/test_i2v_teacher_forcing_context.py:17

Methodapply_ptq

Override the parent method to do nothing, since we need the high-precision weight when doing PTQ with SmoothQuant.

fouroversix/scripts/ptq/evaluators/smoothquant.py:49

Methodapply_ptq

Override the parent method to do nothing, since we need the high-precision weight when calibrating with AWQ.

fouroversix/scripts/ptq/evaluators/awq.py:39

Methodattention_mask

(b, h, q_idx, kv_idx)

wan_5b/modules/causal_model.py:1139

Backward pass for the FP4 linear layer.

fouroversix/src/fouroversix/model/modules/linear.py:78

Backward pass for the FP4 linear layer.

fouroversix/src/fouroversix/model/modules/linear_ori.py:63

Methodbegin_index

The index for the first timestep. It should be set from pipeline with `set_begin_index` method.

wan_5b/utils/fm_solvers.py:209

Methodbegin_index

The index for the first timestep. It should be set from pipeline with `set_begin_index` method.

wan_5b/utils/fm_solvers_unipc.py:144

Methodbfloat16_t>

fouroversix/src/fouroversix/csrc/include/utils.h:57

utils/quant.py:321

Methodbroadcast_tensor_from_root

(self, root_tensor, *, shape)

wan_5b/distributed/sp_training.py:520

Functioncheck_and_get_bf16_matmul_dims

fouroversix/src/fouroversix/csrc/include/element_traits.hpp:80

Functioncheck_and_get_fp4_matmul_dims

fouroversix/src/fouroversix/csrc/include/element_traits.hpp:123

Functioncheck_block_scale_factor_type

fouroversix/src/fouroversix/csrc/include/element_traits.hpp:72

Methodchunk_dict

wan_5b/distributed/sp_training.py:310

Methodchunk_if_needed

(self, tensor, *, dim, already_sharded=False)

wan_5b/distributed/sp_training.py:340

Methodclear_kv_cache

Zero out all tensors in KV cache and cross-attention cache instead of setting them to None. This preserves memory allocation while cl

pipeline/self_forcing_training.py:717

Methodcollect_model_arg

fouroversix/scripts/ptq/evaluators/utils.py:51

Functioncompute_fp4_quant

fouroversix/src/fouroversix/csrc/include/fp4_quant_kernel.h:495

Functioncompute_fp4_quant_block

fouroversix/src/fouroversix/csrc/include/fp4_quant_kernel.h:243

Functioncompute_fp4_quant_prologue

fouroversix/src/fouroversix/csrc/include/fp4_quant_kernel.h:232

Functioncompute_fp4_quant_prologue_block

fouroversix/src/fouroversix/csrc/include/fp4_quant_kernel.h:32

Functioncompute_temporal_freqs

Compute linear temporal RoPE freqs with an optional multi-shot offset.

utils/position_embedding_utils.py:56

wan_5b/distributed/sp_training.py:247

Methodconfigure_torch_compile

( self, *, backend: str = "inductor", mode: str | None = "max-autotune-no-cuda

pipeline/causal_diffusion_inference_sp.py:90

Convert the quantized parameters in gpt oss model to high precision weights.

fouroversix/src/fouroversix/weight_conversions/gpt_oss.py:20

Functionconvert_e2m1_to_fp8_e4m3

(x: torch.Tensor)

fouroversix/src/fouroversix/quantize/quantized_tensor.py:21

Functionconvert_layout_acc_Aregs

fouroversix/src/fouroversix/csrc/include/utils.h:232

Functionconvert_layout_acc_dropout

fouroversix/src/fouroversix/csrc/include/utils.h:254

Functionconvert_layout_acc_rowcol

fouroversix/src/fouroversix/csrc/include/utils.h:219

Methodconvert_noise_to_x0

Convert the diffusion network's noise prediction to x0 predidction. noise: the predicted noise with shape [B, C, H, W] xt: th

utils/scheduler.py:52

← previous next →601–700 of 963, ranked by callers