vllm-omni/docs/user_guide/diffusion_acceleration.md

Diffusion Acceleration Overview

vLLM-Omni supports various cache acceleration methods to speed up diffusion model inference with minimal quality degradation. These methods intelligently cache intermediate computations to avoid redundant work across diffusion timesteps.

Supported Acceleration Methods

vLLM-Omni currently supports two main cache acceleration backends:

1. TeaCache - Hook-based adaptive caching that caches transformer computations when consecutive timesteps are similar
2. Cache-DiT - Library-based acceleration using multiple techniques:
   • DBCache (Dual Block Cache): Caches intermediate transformer block outputs based on residual differences
   • TaylorSeer: Uses Taylor expansion-based forecasting for faster inference
   • SCM (Step Computation Masking): Selectively computes steps based on adaptive masking

Both methods can provide significant speedups (typically 1.5x-2.0x) while maintaining high output quality.

Quick Comparison

Method	Configuration	Description	Best For
TeaCache	cache_backend="tea_cache"	Simple, adaptive caching with minimal configuration	Quick setup, balanced speed/quality
Cache-DiT	cache_backend="cache_dit"	Advanced caching with multiple techniques (DBCache, TaylorSeer, SCM)	Maximum acceleration, fine-grained control

Supported Models

The following table shows which models are currently supported by each cache backend:

Model	Model Identifier	TeaCache	Cache-DiT
Qwen-Image	Qwen/Qwen-Image	✅	✅
Z-Image	Tongyi-MAI/Z-Image-Turbo	❌	✅
Qwen-Image-Edit	Qwen/Qwen-Image-Edit	❌	✅*

* Qwen-Image-Edit support is not fully verified

Performance Benchmarks

The following benchmarks were measured on Qwen/Qwen-Image model generating images with 50 inference steps:

!!! note "Benchmark Disclaimer" These benchmarks are provided for general reference only. The configurations shown use default or common parameter settings and have not been exhaustively optimized for maximum performance. Actual performance may vary based on:

- Specific model and use case
- Hardware configuration
- Careful parameter tuning
- Different inference settings (e.g., number of steps, image resolution)

For optimal performance in your specific scenario, we recommend experimenting with different parameter configurations as described in the detailed guides below.

Configuration	Generation Time	Speedup	Notes
Baseline (diffusers)	20.0s	1.0x	No acceleration
TeaCache (rel_l1_thresh=0.2)	10.47s	1.91x	Recommended default setting
Cache-DiT (DBCache + TaylorSeer)	10.8s	1.85x	Fn=1, Bn=0, W=8, TaylorSeer order=1
Cache-DiT (DBCache + TaylorSeer + SCM)	14.0s	1.43x	Fn=8, Bn=0, W=4, TaylorSeer order=1, SCM fast

Quick Start

Using TeaCache

from vllm_omni import Omni

omni = Omni(
    model="Qwen/Qwen-Image",
    cache_backend="tea_cache",
    cache_config={"rel_l1_thresh": 0.2}  # Optional, defaults to 0.2
)

outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)

Using Cache-DiT

from vllm_omni import Omni

omni = Omni(
    model="Qwen/Qwen-Image",
    cache_backend="cache_dit",
    cache_config={
        "Fn_compute_blocks": 1,
        "Bn_compute_blocks": 0,
        "max_warmup_steps": 8,
        "enable_taylorseer": True,
        "taylorseer_order": 1,
    }
)

outputs = omni.generate(prompt="A cat sitting on a windowsill", num_inference_steps=50)

Documentation

For detailed information on each acceleration method:

• TeaCache Guide - Complete TeaCache documentation, configuration options, and best practices
• Cache-DiT Acceleration Guide - Comprehensive Cache-DiT guide covering DBCache, TaylorSeer, SCM, and configuration parameters