vllm-project/vllm · error · Error

multimodal preprocessing error: {0}

Error message

multimodal preprocessing error: {0}

What it means

EplbState tracks MoE model configuration across all registered models. When a new model is added, its expert-parallel layout (num_routed_experts, num_redundant_experts, num_physical_experts, num_logical_experts, num_expert_groups) must match the first registered model. A mismatch means EPLB cannot share the same physical expert redistribution plan across models.

Source

Thrown at rust/src/chat/src/error.rs:28

#[thiserror_ext(macro(path = "crate::error"))]
pub enum Error {
    #[error("chat request must contain at least one message")]
    EmptyMessages,
    #[error("cannot continue the final message when the last message is not from the assistant")]
    ContinueFinalAssistantWithoutFinalAssistant,
    #[error("chat template is required but none was configured")]
    MissingChatTemplate,
    #[error("chat template error: {0}")]
    ChatTemplate(String),
    #[error("multimodal input is not supported by this chat renderer")]
    UnsupportedMultimodalRenderer,
    #[error("unsupported multimodal content: {0}")]
    UnsupportedMultimodalContent(&'static str),
    #[error("`{modality}` input is not supported by this model")]
    UnsupportedModality { modality: String },
    #[error("At most {limit} {modality}(s) may be provided in one prompt.")]
    MmLimitExceeded { modality: String, limit: usize },
    #[error("multimodal preprocessing error: {0}")]
    Multimodal(#[message] String),
    #[error("{kind} parsing is not available for model `{model_id}`")]
    ParserUnavailableForModel {
        kind: &'static str,
        model_id: String,
    },
    #[error("{kind} parsing is disabled by frontend configuration")]
    ParserDisabled { kind: &'static str },
    #[error(
        "{kind} parser `{name}` is not registered{}",
        available_parser_hint(.available_names)
    )]
    ParserUnavailableByName {
        kind: &'static str,
        name: String,
        available_names: Vec<String>,
    },
    #[error("failed to initialize {kind} parser `{name}`")]

View on GitHub (pinned to c794754062)

Solutions

1. Use models with identical expert-parallel configuration (same routed/redundant/physical/logical expert counts and expert groups)
2. If different models must be served, give each its own EplbState instead of sharing one
3. Re-check num_redundant_experts / EPLB config so both models are built with the same physical expert layout

ref = next(iter(state.model_states.values())).model
same = (ref.num_routed_experts == new.num_routed_experts
        and ref.num_redundant_experts == new.num_redundant_experts
        and ref.num_physical_experts == new.num_physical_experts
        and ref.num_logical_experts == new.num_logical_experts
        and ref.num_expert_groups == new.num_expert_groups)
assert same, 'expert layout differs from already-registered model'

When it happens

Trigger: Registering a second MoE model into EplbState with a different expert count, redundancy, or grouping than an already-registered model — e.g. mixing two different MoE architectures in one EPLB-managed deployment.

Common situations: Serving heterogeneous MoE models under one EPLB state; loading a model with a different redundant-experts setting; config drift after changing eplb redundancy parameters for one model only.

Related errors

• Async EPLB is only supported with the default policy.
• {self.communicator} communicator is incompatible with async
• Configuration error: {0}
• Endpoint not ready after {0}s: {1}
• Backend error: {0}

AI-assisted analysis of vllm-project/vllm@c794754062 (2026-08-14). Data as JSON: /api/errors/842b78892ea688dd. Report an issue: GitHub.