Support New Diffusion Models

Use this guide as a triage flow for finding the smallest change that can support a model. Most new model work should touch a small set of files, even though the runtime is split into separate folders.

Read the Code in This Order

The files are split by runtime responsibility. For a new model, read the request path first:

1. registry.py chooses the model family, sampling params, and pipeline config.
2. configs/pipeline_configs/{model}.py defines model-specific denoising and decoding behavior.
3. runtime/pipelines/{model}.py wires modules into stages.
4. runtime/pipelines_core/stages/ runs the shared stage logic.
5. runtime/models/ contains native model components only when the architecture cannot be reused.

That is the dependency direction. Avoid making a model PR that requires readers to jump between folders in a different order.

runtime/models/ owns modeling code: checkpoint-defined neural modules, architecture wrappers, and weight-loading or forward-path details that are intrinsic to one model family. Reusable serving infrastructure belongs in SGLang-Diffusion runtime folders such as runtime/cache/, runtime/distributed/, runtime/utils/, or shared pipeline stages. This includes cache managers, graph runners, process-group transport, request utilities, and common action-policy helpers. Model packages may call these helpers. Keep ownership in shared runtime folders unless the code is truly architecture-specific.

Start With the Smallest Change

Before adding files, decide which path fits the model.

Do not add a folder just to mirror the Diffusers repository layout. Add a new file only when an existing pipeline, stage, module, config, or sampler cannot express the behavior clearly.

Minimal File Map

The source tree is split by runtime responsibility. That split is useful for optimization. Keep new model PRs focused on the files required by model behavior.

For a new native architecture, the common minimum is:

1. registry.py
2. configs/sample/{model}.py
3. configs/pipeline_configs/{model}.py
4. runtime/pipelines/{model}.py
5. runtime/models/dits/{model}.py

Every extra file should map to model behavior that existing code cannot express clearly.

For dVLA or other non-image diffusion policies, keep the same ownership rule. The policy network, VLM/action expert modules, checkpoint mapping, and model-specific forward code belong under runtime/models/. Prefix caches, request-local contexts, denoising graph runners, OpenPI-compatible transport, and prefix/action process-group utilities should be shared SGLang-Diffusion runtime infrastructure when they are useful beyond the first model.

Read the Reference First

Use the model's Diffusers pipeline, official implementation, or model_index.json as the source of truth. Write down:

• Which modules must be loaded: tokenizer, text encoder, image encoder, transformer, scheduler, VAE, processor, and any extra adapters.
• The prompt and image encoding flow.
• Latent shape, packing, scale, shift, dtype, and device rules.
• Timestep and sigma schedule.
• The exact forward() kwargs expected by the denoising network.
• VAE decode rules and output post-processing.

If the new model is close to Flux, Qwen-Image, GLM-Image, Wan, HunyuanVideo, or LTX, extend that implementation before starting from an empty file.

Choose a Pipeline Shape

SGLang-Diffusion uses ComposedPipelineBase to wire stages together. Most native pipelines should use one of these two shapes.

Prefer standard stages when possible. Use a model-specific BeforeDenoisingStage when trying to force the model into shared stages would create many conditionals.

Implement the Pieces

1. Sampling Params

Create request parameters only for values users can set at runtime.

# python/sglang/multimodal_gen/configs/sample/my_model.py
from dataclasses import dataclass

from sglang.multimodal_gen.configs.sample.sampling_params import ImageSamplingParams


@dataclass
class MyModelSamplingParams(ImageSamplingParams):
    guidance_scale: float = 4.0
    num_inference_steps: int = 28

2. Pipeline Config

PipelineConfig is where shared denoising and decoding stages get model-specific callbacks.

# python/sglang/multimodal_gen/configs/pipeline_configs/my_model.py
from dataclasses import dataclass, field


@dataclass
class MyModelPipelineConfig(ImagePipelineConfig):
    task_type: ModelTaskType = ModelTaskType.T2I
    should_use_guidance: bool = True
    dit_config: DiTConfig = field(default_factory=MyModelDiTConfig)
    vae_config: VAEConfig = field(default_factory=MyModelVAEConfig)

    def prepare_pos_cond_kwargs(self, batch, latent_model_input, t, **kwargs):
        return {
            "hidden_states": latent_model_input,
            "encoder_hidden_states": batch.prompt_embeds[0],
            "timestep": t,
        }

    def prepare_neg_cond_kwargs(self, batch, latent_model_input, t, **kwargs):
        return {
            "hidden_states": latent_model_input,
            "encoder_hidden_states": batch.negative_prompt_embeds[0],
            "timestep": t,
        }

Make these kwargs match the denoising module's forward() signature exactly.

3. Pipeline Wiring

Use the standard helper when the model fits it.

# python/sglang/multimodal_gen/runtime/pipelines/my_model.py
class MyModelPipeline(LoRAPipeline, ComposedPipelineBase):
    pipeline_name = "MyModelPipeline"

    _required_config_modules = [
        "text_encoder",
        "tokenizer",
        "transformer",
        "scheduler",
        "vae",
    ]

    def create_pipeline_stages(self, server_args: ServerArgs):
        self.add_standard_t2i_stages()


EntryClass = [MyModelPipeline]

Use a model-specific pre-processing stage when the reference pipeline cannot be cleanly expressed by standard helpers.

class MyModelPipeline(LoRAPipeline, ComposedPipelineBase):
    pipeline_name = "MyModelPipeline"

    _required_config_modules = [
        "text_encoder",
        "tokenizer",
        "transformer",
        "scheduler",
        "vae",
    ]

    def create_pipeline_stages(self, server_args: ServerArgs):
        self.add_stage(
            MyModelBeforeDenoisingStage(
                text_encoder=self.get_module("text_encoder"),
                tokenizer=self.get_module("tokenizer"),
                transformer=self.get_module("transformer"),
                scheduler=self.get_module("scheduler"),
                vae=self.get_module("vae"),
            )
        )
        self.add_stage(
            DenoisingStage(
                transformer=self.get_module("transformer"),
                scheduler=self.get_module("scheduler"),
            )
        )
        self.add_standard_decoding_stage()


EntryClass = [MyModelPipeline]

4. Optional Before-Denoising Stage

A BeforeDenoisingStage should populate the batch fields consumed by DenoisingStage.

class MyModelBeforeDenoisingStage(PipelineStage):
    @torch.no_grad()
    def forward(self, batch: Req, server_args: ServerArgs) -> Req:
        prompt_embeds, negative_prompt_embeds = self._encode_prompt(batch)
        latents = self._prepare_latents(batch)
        timesteps, sigmas = self._prepare_timesteps(batch)

        batch.prompt_embeds = [prompt_embeds]
        batch.negative_prompt_embeds = [negative_prompt_embeds]
        batch.latents = latents
        batch.timesteps = timesteps
        batch.num_inference_steps = len(timesteps)
        batch.sigmas = sigmas.tolist()
        batch.raw_latent_shape = latents.shape
        return batch

Required fields for DenoisingStage:

5. Denoising Module

Add a file under runtime/models/dits/ only when the architecture is new. Reuse existing encoders, VAEs, schedulers, normalization layers, and fused kernels whenever possible.

For multi-GPU serving, add TP/SP support after the single-GPU path is correct. Useful references:

• runtime/models/dits/wanvideo.py for TP plus SP.
• runtime/models/dits/qwen_image.py for USP attention.

6. Registry

Register the family once the sampling params and pipeline config exist.

register_configs(
    model_family="my_model",
    sampling_param_cls=MyModelSamplingParams,
    pipeline_config_cls=MyModelPipelineConfig,
    hf_model_paths=["org/my-model"],
)

The pipeline file is discovered through its EntryClass; do not add a second pipeline registry unless the existing registry requires it.

Verify the Port

Use one deterministic prompt and seed while comparing with the reference implementation.

1. Run a single-GPU smoke test and check that the output contains coherent content.
2. Compare latent scale and shift, timestep order, sigma values, and conditioning kwargs against Diffusers or the official implementation.
3. Verify VAE decode and post-processing separately from denoising.
4. If the model supports LoRA, CFG parallelism, TP, SP, or disaggregation, test each feature explicitly.
5. Add or update docs, examples, or the compatibility matrix when users need a new launch command.

Common failure points:

• Wrong latent scale or shift.
• Reversed or dtype-mismatched timesteps.
• Missing negative embeddings when CFG is enabled.
• Conditioning kwarg names mismatched with the DiT forward().
• Rotary embedding shape or style mismatch.
• Decoding packed latents without restoring raw_latent_shape.

PR Checklist

• Reused an existing family, stage, module, scheduler, or VAE wherever possible.
• Kept the new-model touch surface small and justified any extra files.
• Added SamplingParams, PipelineConfig, pipeline wiring, DiT module, and registry entry when native support is needed.
• Confirmed pipeline_name matches the Diffusers model_index.json _class_name when applicable.
• Confirmed _required_config_modules matches the model repo.
• Verified image or video quality against a reference output.
• Tested multi-GPU paths if the PR claims TP, SP, CFG parallelism, or distributed serving support.