hpcaitech/Open-Sora · error · ValueError

Didn't get conditional input for conditional model.

Error message

Didn't get conditional input for conditional model.

What it means

When the MMDiT config enables cond_embed (conditional-input projection), prepare_block_inputs requires a non-None cond tensor to add via self.cond_in(cond). Passing no cond (or cond=None) means the conditional branch cannot execute, so it fails fast.

Source

Thrown at opensora/models/mmdit/model.py:179

        y_vec: Tensor,  # clip encoded vec
        cond: Tensor = None,
        guidance: Tensor | None = None,
    ):
        """
        obtain the processed:
            img: projected noisy img latent,
            txt: text context (from t5),
            vec: clip encoded vector,
            pe: the positional embeddings for concatenated img and txt
        """
        if img.ndim != 3 or txt.ndim != 3:
            raise ValueError("Input img and txt tensors must have 3 dimensions.")

        # running on sequences img
        img = self.img_in(img)
        if self.config.cond_embed:
            if cond is None:
                raise ValueError("Didn't get conditional input for conditional model.")
            img = img + self.cond_in(cond)

        vec = self.time_in(timestep_embedding(timesteps, 256))
        if self.config.guidance_embed:
            if guidance is None:
                raise ValueError(
                    "Didn't get guidance strength for guidance distilled model."
                )
            vec = vec + self.guidance_in(timestep_embedding(guidance, 256))
        vec = vec + self.vector_in(y_vec)

        txt = self.txt_in(txt)

        # concat: 4096 + t*h*2/4
        ids = torch.cat((txt_ids, img_ids), dim=1)
        pe = self.pe_embedder(ids)

        if self._input_requires_grad:

View on GitHub (pinned to 7ad6a96a13)

Solutions

1. Pass cond=... (the conditional embedding tensor, shape compatible with img sequence length) to forward
2. If you intended an unconditional model, set config.cond_embed=False (and use a matching checkpoint)
3. Audit the call site: mmdit_model_forward and ckpt variants need the same argument

Example fix

# before
out = model(img, txt, timesteps, y_vec=y_vec)  # cond_embed=True
# after
out = model(img, txt, timesteps, y_vec=y_vec, cond=cond_emb)

if model.config.cond_embed:
    assert cond is not None, "cond_embed=True requires the cond tensor"

def needs_cond(model) -> bool:
    return bool(getattr(model.config, "cond_embed", False))

try:
    out = model(img, txt, t, y_vec=y_vec, cond=cond)
except ValueError as e:
    if "conditional input" in str(e):
        raise TypeError("checkpoint is conditional; supply cond= or use a cond_embed=False checkpoint") from e
    raise

When it happens

Trigger: Instantiating the model with config.cond_embed=True and calling forward without the cond argument (all forward variants: mmdit_model_forward, forward_ckpt, forward_selective_ckpt).

Common situations: Loading a conditional checkpoint but running an unconditional-generation code path; shared inference scripts that omit cond for unconditional models; refactor dropping the cond kwarg.

Understand the failure class

Background: "missing required argument" and "the following required arguments were not provided": what required-argument errors mean and how to fix them — this error's family across 20 libraries.

Related errors

• Didn't get guidance strength for guidance distilled model.
• Hidden size {config.hidden_size} must be divisible by num_he
• Got {config.axes_dim} but expected positional dim {pe_dim}
• Input img and txt tensors must have 3 dimensions.
• ContextParallelAttention should not be initialized directly.

AI-assisted analysis of hpcaitech/Open-Sora@7ad6a96a13 (2026-08-28). Data as JSON: /api/errors/f4b6305801f81ce7. Report an issue: GitHub.