sgl-project/sglang · error · ValueError

Expected {seq_len=} > 0 for packed token latents.

Error message

Expected {seq_len=} > 0 for packed token latents.

What it means

LTX-2's _infer_video_latent_frames_and_tokens_per_frame needs a strictly positive packed-token sequence length to reconstruct latent frame geometry for SP sharding; seq_len <= 0 would divide nowhere and indicates an empty/corrupt latent.

Source

Thrown at python/sglang/multimodal_gen/configs/pipeline_configs/ltx_2.py:336

        # kernel runs and move bf16 output. The fp8 path makes its own copy.
        return latents

    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 "

View on GitHub (pinned to 0132848349)

Solutions

1. Verify video dimensions produce at least one latent token: frames >= temporal stride, height/width large enough after VAE downsampling
2. Check the packed latent tensor shape right before sharding and abort with a clear message if seq dimension is 0
3. Fix upstream packing/truncation so the token dimension is non-empty

Example fix

# before
latents = pack(latent)  # may be empty for tiny inputs
shard_latents_for_sp(latents)

# after
assert latents.shape[-2] > 0, f"empty packed latents: {latents.shape}"
shard_latents_for_sp(latents)

seq = packed_latents.shape[-2]
assert seq > 0, f"packed latent token seq_len must be > 0, got {seq}"

except ValueError as e:
    if "seq_len" in str(e):
        # re-check dims/frames and resubmit with valid resolution/frames
        raise ValueError(f"invalid video geometry: {height}x{width}x{frames}") from e

When it happens

Trigger: Calling shard_latents_for_sp with a packed latent whose token sequence length is 0 or negative — e.g. an all-padding latent, a zero-frame video, or an upstream packing bug producing an empty token dimension.

Common situations: Passing height/width/frame values that round down to zero latent tokens; a VAE encode returning an empty tensor; truncation/stride settings that eliminate all frames.

Related errors

• LTX-2 SP time-sharding for packed token latents currently re
• Krea-2 sequence parallelism does not support ragged/padded m
• padding_side must be 'left' or 'right', got {padding_side}
• Unsupported text encoder output: expected `hidden_states`.
• num_inference_steps must be positive, got {steps}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/abb6def915343368. Report an issue: GitHub.