pub fn load_dflash_weights(
drafter_store: &WeightStore,
drafter_config: &DflashConfig,
_gpu: &dyn GpuBackend,
_tp_size: usize,
) -> Result<Option<DflashWeights>>Expand description
2026-09-25: Load the drafter’s weights from the drafter’s own
WeightStore.
Returns Ok(None) when store_has_dflash_weights is false. Otherwise
fc, the two model norms and num_hidden_layers layers of norms and
projections are required; the conv, selector, Markov and confidence
tensors are optional as their fields describe. _tp_size is unused, so
every rank loads the whole drafter.