sgl-project/sglang · error · ValueError

Invalid {self.vae_scale_factor=}. Must be > 0.

Error message

Invalid {self.vae_scale_factor=}. Must be > 0.

What it means

LTX-2 pipeline config validates that vae_scale_factor is a positive integer before computing latent spatial dims for sequence-parallel (SP) time-sharding of packed token latents. The latent height/width are derived as batch.height // vae_scale_factor, so a zero or negative scale factor would produce garbage dimensions. This ValueError fires when the configured VAE scale factor is <= 0.

Source

Thrown at python/sglang/multimodal_gen/configs/pipeline_configs/ltx_2.py:338

    def _infer_video_latent_frames_and_tokens_per_frame(
        self, batch, seq_len: int
    ) -> tuple[int, int]:
        """Infer latent-frame count and tokens-per-frame for packed token latents [B, S, D].

        Notes:
        - This assumes `patch_size_t == 1` (no temporal patching).
        - Tokens are ordered as (frame, height, width) after packing.
        """
        if int(self.patch_size_t) != 1:
            raise ValueError(
                "LTX-2 SP time-sharding for packed token latents currently requires "
                f"{self.patch_size_t=}. (Expected 1)"
            )
        if int(seq_len) <= 0:
            raise ValueError(f"Expected {seq_len=} > 0 for packed token latents.")
        if int(self.vae_scale_factor) <= 0:
            raise ValueError(f"Invalid {self.vae_scale_factor=}. Must be > 0.")
        if int(self.patch_size) <= 0:
            raise ValueError(f"Invalid {self.patch_size=}. Must be > 0.")

        latent_height = int(batch.height) // int(self.vae_scale_factor)
        latent_width = int(batch.width) // int(self.vae_scale_factor)
        if latent_height <= 0 or latent_width <= 0:
            raise ValueError(
                "Invalid latent H/W computed from batch.height/width: "
                f"{batch.height=} {batch.width=} {self.vae_scale_factor=}"
            )
        if (latent_height % int(self.patch_size)) != 0 or (
            latent_width % int(self.patch_size)
        ) != 0:
            raise ValueError(
                "Invalid spatial patching for packed token latents. Expected latent H/W "
                "to be divisible by patch_size, got "
                f"{latent_height=} {latent_width=} {self.patch_size=}."
            )

View on GitHub (pinned to 0132848349)

Solutions

1. Check the config source and set vae_scale_factor to the model's correct value (LTX-2 typically uses 8 for spatial VAE downsampling)
2. If loading from a checkpoint/config dict, verify the key exists and is parsed as a positive int before constructing the pipeline config
3. Add a validation step at config construction time so the error surfaces earlier with more context

Example fix

// before
config = LTX2PipelineConfig.from_dict({"vae_scale_factor": 0, ...})
batch = ...
config.shard_latents_for_sp(batch)  # ValueError

// after
config = LTX2PipelineConfig.from_dict({"vae_scale_factor": 8, ...})
batch = ...
config.shard_latents_for_sp(batch)

assert isinstance(config.vae_scale_factor, int) and config.vae_scale_factor > 0, config.vae_scale_factor

When it happens

Trigger: Calling shard_latents_for_sp (which calls _infer_video_latent_frames_and_tokens_per_frame) with a pipeline config whose vae_scale_factor was set to 0 or a negative number, typically from a malformed config file or override.

Common situations: Custom LTX-2 pipeline configs edited by hand, configs generated from a checkpoint missing the VAE metadata (defaulting to 0), or programmatic overrides that accidentally assign 0.

Understand the failure class

Background: "Invalid value" and "allowed values are" config errors: what your library rejected and how to fix it — this error's family across 41 libraries.

Related errors

• Invalid {self.patch_size=}. Must be > 0.
• Invalid latent H/W computed from batch.height/width: {batch.
• Invalid tokens_per_frame={tokens_per_frame} from {latent_hei
• LTX-2 token latents seq_len={seq_len} is not divisible by to
• LTX-2 SP time-sharding for packed token latents currently re

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/24c4cf613b1f8afa. Report an issue: GitHub.