Optimizations¶

This page describes the various options for speeding up generation times in FastVideo.

Table of Contents¶

• Optimized Attention Backends

  • Flash Attention
  • Sliding Tile Attention (Archived)
  • Sage Attention
  • Sage Attention 3

• FP8 Weight Quantization

• Adaptive Guidance (CFG gating)

• torch.compile

Attention Backends¶

Available Backends¶

• Torch SDPA: FASTVIDEO_ATTENTION_BACKEND=TORCH_SDPA
• Flash Attention 2 and 3: FASTVIDEO_ATTENTION_BACKEND=FLASH_ATTN
• Video Sparse Attention: FASTVIDEO_ATTENTION_BACKEND=VIDEO_SPARSE_ATTN
• Sage Attention: FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN
• Sage Attention 3: FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN_THREE
• Attn-QAT inference (modified SageAttention3 FP4, sm_120/RTX 5090): FASTVIDEO_ATTENTION_BACKEND=ATTN_QAT_INFER
• Video MoBA Attention: FASTVIDEO_ATTENTION_BACKEND=VMOBA_ATTN
• Sparse Linear Attention: FASTVIDEO_ATTENTION_BACKEND=SLA_ATTN
• SageSLA Attention: FASTVIDEO_ATTENTION_BACKEND=SAGE_SLA_ATTN
• Sliding Tile Attention (archived branch only): FASTVIDEO_ATTENTION_BACKEND=SLIDING_TILE_ATTN

Configuring Backends¶

There are two ways to configure the attention backend in FastVideo.

1. In Python¶

In python, set the FASTVIDEO_ATTENTION_BACKEND environment variable before instantiating VideoGenerator like this:

os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "VIDEO_SPARSE_ATTN"

2. In CLI¶

You can also set the environment variable on the command line:

FASTVIDEO_ATTENTION_BACKEND=SAGE_ATTN python example.py

Flash Attention¶

FLASH_ATTN

We recommend always installing Flash Attention 2:

uv pip install flash-attn==2.7.4.post1 --no-build-isolation

And if using a Hopper+ GPU (ie H100), installing Flash Attention 3 by compiling it from source (takes about 10 minutes for me):

git clone https://github.com/Dao-AILab/flash-attention.git && cd flash-attention

cd hopper
uv pip install ninja
python setup.py install

Flash Attention 4 (opt-in)¶

FastVideo never auto-selects FlashAttention-4 (flash_attn.cute) just because it is installed: its CuTeDSL kernels JIT-compile per shape family and can fail at runtime on some GPU/shape combinations. To use FA4, install the pinned flash-attn-4 build (see the flash-attn-4 source in pyproject.toml) and set:

export FASTVIDEO_FA4=1

On GPUs below sm90 a capability gate routes to FlashAttention-2 the calls FA4 cannot serve there: grad-enabled (training) attention (FA4's backward requires sm90+) and GQA attention (FA4's pack_gqa fails to JIT-compile below sm90). On sm90+ both run on FA4. If FA4 is unusable while FASTVIDEO_FA4=1 is set, FastVideo fails loudly instead of silently falling back.

FP4 Flash Attention 4 (Blackwell only)¶

FLASH_ATTN with --nvfp4_fa4

On Blackwell GPUs (B200/B300), you can enable FP4 quantized Q/K attention for up to 1.31x kernel speedup over BF16 FA4, peaking at 2018 TFLOPS. This quantizes Q and K to NVFP4 E2M1 with per-block E4M3 scale factors while keeping V in BF16 or FP8.

See the Attn-QAT paper and flash-attention-fp4 benchmark results for details.

Requirements¶

• GPU: NVIDIA Blackwell (sm100a or sm103a) — B200, B300, GB200, GB300
• CUDA: 13.0+
• Python: 3.10 or 3.11

Installation¶

Install the FP4 flash attention kernel (without upgrading your existing torch):

# branch fix/cutlass-dsl-4.5 carries the cutlass-dsl 4.5 fix (cute.core.ThrMma
# -> cute.ThrMma); switch back to @fp4 once hao-ai-lab/flash-attention-fp4#2 merges.
pip install --no-deps "git+ssh://git@github.com/hao-ai-lab/flash-attention-fp4.git@fix/cutlass-dsl-4.5#subdirectory=flash_attn/cute"
pip install "nvidia-cutlass-dsl>=4.5.2" apache-tvm-ffi flashinfer-python

The --no-deps flag prevents upgrading torch/torchvision. Use the supported PyTorch 2.12.0 and CUDA 13 environment for this kernel.

Usage¶

Enable FP4 attention via the --nvfp4_fa4 flag:

python examples/inference/optimizations/fp4_attn_wan2_1_1_3b.py --nvfp4_fa4

Or in Python via the nvfp4_fa4 kwarg (sets env vars automatically):

from fastvideo import VideoGenerator
gen = VideoGenerator.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    nvfp4_fa4=True,
    num_gpus=1,
    use_fsdp_inference=False,  # FSDP is incompatible with FP4 pointer path
)
gen.generate_video(prompt="A raccoon in sunflowers", save_video=True)

Known Limitations¶

• use_fsdp_inference=True is incompatible with the FP4 path (FSDP shards invalidate tensor pointers)
• Per-call cosine similarity vs BF16: ~0.99 (slight quantization error accumulates over denoising steps)
• Only supports headdim >= 128

NVFP4 + Attn-QAT (modified SageAttention3, Blackwell sm_120)¶

ATTN_QAT_INFER with transformer_quant=nvfp4_qat

Runs the DiT fully in 4-bit: NVFP4 linear layers (activations quantized on the fly) plus the modified SageAttention3 FP4 attention backend. This is the inference half of the Quantization-Aware Distillation (QAD) recipe and the path used for the RTX 5090 release.

The attn_qat_infer kernel hard-gates on sm_120 (consumer Blackwell / RTX 5090); on other GPUs the backend logs a notice and falls back to Flash Attention. See the Attn-QAT paper.

Enable both halves — attention via the env var, linear via transformer_quant:

import os
os.environ["FASTVIDEO_ATTENTION_BACKEND"] = "ATTN_QAT_INFER"

from fastvideo import VideoGenerator
from fastvideo.layers.quantization import get_quantization_config
gen = VideoGenerator.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    num_gpus=1,
    # Wan-2.1 uses the nvfp4_qat config (NVFP4 is LTX2-specific). Pass an
    # instance — the bare string is not resolved on the from_pretrained path.
    transformer_quant=get_quantization_config("nvfp4_qat")(),
    use_fsdp_inference=False,     # FSDP shards invalidate the FP4 tensor pointers
)
gen.generate(request={"prompt": "A raccoon in sunflowers", "output": {"save_video": True}})

Or run the example script:

python examples/inference/optimizations/nvfp4_qat_wan2_1_1_3b.py
python examples/inference/optimizations/nvfp4_qat_wan2_1_1_3b.py --bf16  # baseline

Sliding Tile Attention (Archived)¶

SLIDING_TILE_ATTN

The full STA integration in fastvideo/ is archived from main and preserved at:

• https://github.com/hao-ai-lab/FastVideo/tree/sta_do_not_delete

We keep STA off main because we believe VSA is strictly better than STA for the actively maintained FastVideo path.

Kernel code in fastvideo-kernel is still retained. For mask search and STA inference workflow, see STA docs.

Video Sparse Attention¶

VIDEO_SPARSE_ATTN

Video Sparse Attention is provided by fastvideo-kernel. See VSA docs for installation details.

Sage Attention¶

SAGE_ATTN

To use SageAttention 2.1.1, please compile from source:

git clone https://github.com/thu-ml/SageAttention.git
cd sageattention
python setup.py install  # or uv pip install -e .

Sage Attention 3¶

SAGE_ATTN_THREE

SageAttention 3 is an advanced attention mechanism that leverages FP4 quantization and Blackwell GPU Tensor Cores for significant performance improvements.

Hardware Requirements¶

• RTX5090

Installation¶

Note that Sage Attention 3 requires python>=3.13, torch>=2.8.0, and CUDA 13. If you are using uv and torch==2.8.0, make sure that sentencepiece==0.2.1 in the pyproject.toml file.

To use Sage Attention 3 in FastVideo, follow the README.md in the linked repository to install the package from source.

V-MoBA / SLA / SageSLA¶

These backends are model-specific and require the corresponding kernels and dependencies. Use the support matrix and model examples to confirm compatibility before enabling them.

FP8 Weight Quantization¶

transformer_quant="FP8"

Quantizes DiT linear layers (attention projections and FFN) to FP8 e4m3.

On GPUs older than sm89, the FP8 matmul falls back to a bf16 dequant path automatically.

Requirements¶

• GPU: sm89+ (H100, L40S, RTX 4090, or newer) for hardware FP8 compute
• No additional packages required beyond the base FastVideo install

Usage¶

from fastvideo import VideoGenerator
from fastvideo.layers.quantization import get_quantization_config

gen = VideoGenerator.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    # Pass an instance — the bare string is not resolved on the from_pretrained path.
    transformer_quant=get_quantization_config("FP8")(),          # per-tensor (default)
    # transformer_quant=get_quantization_config("FP8")(granularity="channel"),  # slower, higher accuracy
)
gen.generate(request={"prompt": "A raccoon in sunflowers", "output": {"save_video": True}})

Or run the example script:

python examples/inference/optimizations/fp8_wan2_1_1_3b.py
python examples/inference/optimizations/fp8_wan2_1_1_3b.py --granularity channel
python examples/inference/optimizations/fp8_wan2_1_1_3b.py --bf16  # baseline

Granularity¶

torch.compile¶

FastVideo can torch.compile the DiT (transformer) for a substantial end-to-end speedup. It is off by default and enabled per-run.

Enabling¶

generator = VideoGenerator.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    enable_torch_compile=True,
)

A complete A/B example (eager vs compiled, warmup excluded) is in examples/inference/optimizations/torch_compile_example.py.

fastvideo generate is config-file driven; to enable torch.compile from the CLI, set the relevant field in your run config and pass it via fastvideo generate --config run.yaml. There is no top-level --enable-torch-compile flag on the subcommand.

Only DiT submodules that declare _compile_conditions are compiled (most shipped models). The text encoder and VAE are not compiled by this flag.

What to expect¶

The speedup is configuration-dependent — it varies with model, resolution, step count, and GPU. Treat the number above as one measured data point, not a guarantee; benchmark your own config (recipe below).

There is a one-time graph-build cost on the first generation (tens of seconds to minutes, model-dependent). It amortizes over subsequent generations with the same input shapes. Always exclude the first (warmup) generation when measuring steady-state latency — measuring the warmup is the most common way to wrongly conclude "compile is slower".

Numerics. Inductor's lowering is designed to preserve eager semantics within floating-point tolerance, but per-model equivalence is not asserted by any standing SSIM regression here — the SSIM tests in fastvideo/tests/ssim/ run with enable_torch_compile disabled. If you depend on compile output staying close to eager (or your previous compiled run), run an MS-SSIM gate on your config, especially when combining enable_torch_compile=True with other numerics-affecting flags (quantized attention backends, FP4, layerwise offload edge cases).

Known interactions¶

• Layerwise CPU offload (dit_layerwise_offload=True, the default): the offload hook previously caused an implicit graph break once per transformer layer, fragmenting the compiled region. Addressed in hao-ai-lab/FastVideo#1365 — keep that fix to get a clean compiled region under the default offload path.
• mode="reduce-overhead" / CUDA graphs: not yet supported end-to-end. The attention dispatch is an untraceable custom op and still breaks the graph, which CUDA-graph trees cannot span. Use the default inductor mode (shown above) until that is resolved.

Extra torch.compile options are passed through torch_compile_kwargs (a dict), accepted by VideoGenerator.from_pretrained(...) and by the CLI as a JSON string via --torch-compile-kwargs. Example (currently not recommended — see the CUDA-graphs caveat above):

VideoGenerator.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    enable_torch_compile=True,
    torch_compile_kwargs={"mode": "reduce-overhead"},  # may error today
)

Benchmarking torch.compile¶

Same discipline as attention backends — same prompt, same seed, same config; discard the first generation (graph build):

import time
from fastvideo import VideoGenerator

gen = VideoGenerator.from_pretrained("your-model-id", enable_torch_compile=True)
req = {"prompt": "Your prompt", "sampling": {"seed": 1024},
       "output": {"save_video": False}}
gen.generate(req)                                            # warmup: graph build, discard
t0 = time.perf_counter()
gen.generate(req)                                            # measured: shapes reused
print(f"compiled steady-state: {time.perf_counter() - t0:.2f}s")

See examples/inference/optimizations/torch_compile_example.py for a baseline-vs-compile A/B with the warmup correctly excluded.

Benchmarking different optimizations¶

To benchmark backend performance, generate the same prompt with the same seed and compare end-to-end generation times:

import os
import time

for backend in ["TORCH_SDPA", "FLASH_ATTN", "SAGE_ATTN"]:
    os.environ["FASTVIDEO_ATTENTION_BACKEND"] = backend
    generator = VideoGenerator.from_pretrained("your-model-id")
    start_time = time.perf_counter()
    generator.generate_video(
        prompt="Your prompt",
        seed=1024,
    )
    elapsed = time.perf_counter() - start_time
    print(f"{backend}: {elapsed:.2f}s")

Note: reinstantiate VideoGenerator after changing FASTVIDEO_ATTENTION_BACKEND.

Adaptive Guidance (CFG gating)¶

CFG gating accelerates classifier-free guidance by reusing the cached noise_pred_cond - noise_pred_uncond delta after a configurable fraction of the denoising schedule, skipping the unconditional model forward for the remaining steps. The technique is the LinearAG variant of Adaptive Guidance (Castillo et al. 2023, arXiv:2312.12487).

Enabling¶

Set the FASTVIDEO_CFG_GATE_STEP environment variable to a float in [0, 1]:

export FASTVIDEO_CFG_GATE_STEP=0.5

Trade-offs¶

• Memory: one extra model-output-sized tensor per rank held during the gating window.
• Quality: VBench-measured quality is preserved within noise on 4 of 5 dimensions at FASTVIDEO_CFG_GATE_STEP=0.5 for Wan T2V 1.3B per the PR's reported numbers (see #1372).
• Speed: ~22% e2e on 4xL40S and ~24% on 1xH100 at the same settings.

Default behavior is byte-for-byte equivalent to the legacy two-pass CFG path; the feature is fully opt-in.