huggingface/transformers · error · ValueError

Calling `get_mtp_config` on a config containing `layer_types

Error message

Calling `get_mtp_config` on a config containing `layer_types` without `mtp_layer_types` is ambiguous

What it means

ValueError from get_mtp_config: the text config defines layer_types (per-layer attention type list) but not mtp_layer_types. MTP layers restart layer indexing at 0, so the code must replace layer_types with an MTP-specific list; without mtp_layer_types it cannot know which layer types the MTP layers use.

Source

Thrown at src/transformers/configuration_utils.py:1402

        the indexing of layers at 0).
        """
        # Start from the text config
        text_config = copy.deepcopy(self.get_text_config(decoder=True))
        num_mtp_layers = getattr(text_config, "num_mtp_layers", None)
        # In this case, raise
        if num_mtp_layers is None:
            raise ValueError("Calling `get_mtp_config` on a config without `num_mtp_layers`")

        layer_types = getattr(text_config, "layer_types", None)
        mtp_layer_types = getattr(text_config, "mtp_layer_types", None)
        mlp_layer_types = getattr(text_config, "mlp_layer_types", None)
        mtp_mlp_layer_types = getattr(text_config, "mtp_mlp_layer_types", None)

        # Replace the index-based fields so that we can call `XXXDecoderLayer(config, layer_idx)` with the mtp config, and create
        # the correct layers
        if layer_types is not None:
            if mtp_layer_types is None:
                raise ValueError(
                    "Calling `get_mtp_config` on a config containing `layer_types` without `mtp_layer_types` is ambiguous"
                )
            text_config.layer_types = mtp_layer_types
        if mlp_layer_types is not None:
            if mtp_mlp_layer_types is None:
                raise ValueError(
                    "Calling `get_mtp_config` on a config containing `mlp_layer_types` without `mtp_mlp_layer_types` is ambiguous"
                )
            text_config.mlp_layer_types = mtp_mlp_layer_types

        # This is needed to be correct in several places, e.g. when creating the cache
        text_config.num_hidden_layers = num_mtp_layers

        return text_config


def get_configuration_file(configuration_files: list[str]) -> str:
    """

View on GitHub (pinned to a597f97485)

Solutions

  1. Set text_config.mtp_layer_types to a list whose length equals num_mtp_layers, e.g. ["dense"] * num_mtp_layers.
  2. Copy an official MTP-capable checkpoint's config (e.g. DeepSeek-V3) as the template for these fields.
  3. If the model genuinely has no MTP layers, remove num_mtp_layers instead of leaving it set without mtp fields.

Example fix

// before
config.layer_types = ["dense", "linear"] * 16
config.num_mtp_layers = 1
config.get_mtp_config()  # ValueError

// after
config.mtp_layer_types = ["dense"]
config.get_mtp_config()
Defensive patterns

Strategy: validation

Validate before calling

text_cfg = config.get_text_config(decoder=True)
if getattr(text_cfg, "num_mtp_layers", None) is not None:
    if getattr(text_cfg, "layer_types", None) is not None and getattr(text_cfg, "mtp_layer_types", None) is None:
        text_cfg.mtp_layer_types = ["dense"] * text_cfg.num_mtp_layers  # or appropriate types

Type guard

def mtp_layer_types_are_consistent(text_cfg) -> bool:
    return getattr(text_cfg, "layer_types", None) is None or getattr(text_cfg, "mtp_layer_types", None) is not None

Prevention

When it happens

Trigger: A config with layer_types=["dense",...] and num_mtp_layers>0 but no mtp_layer_types attribute, then get_mtp_config() is called (typically during DeepSeek-V3/MTP-style model construction).

Common situations: Hand-written configs for hybrid-attention models that add MTP layers without filling in the MTP fields; older converted checkpoints predating the mtp_layer_types field.

Related errors

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/a59d68dc62feef65. Report an issue: GitHub.