Struct MixedBatchResult 

pub struct MixedBatchResult {
    pub decode_logits: DevicePtr,
    pub prefill_logits: Vec<DevicePtr>,
}

Result of a fully-batched mixed forward pass: M decode tokens + N prefill chunks in one pass.

Fields

decode_logits: DevicePtr

Logits for decode lanes: [M, vocab] BF16. NULL if no decode lanes.

prefill_logits: Vec<DevicePtr>

Logits per prefill stream — one DevicePtr per stream in the input slice, in the same order. Each entry is [1, vocab] BF16 when that stream’s chunk was is_last_chunk, or NULL otherwise.