bridge.training.utils.packed_seq_utils

Module Contents

Functions

get_packed_seq_params

Build packed sequence parameters from a batch dictionary.

API

bridge.training.utils.packed_seq_utils.get_packed_seq_params(

batch: dict[str, torch.Tensor],

) → megatron.core.packed_seq_params.PackedSeqParams

Build packed sequence parameters from a batch dictionary.

The function squeezes possible batch dimensions and removes any padding marked by -1 values. It returns a PackedSeqParams instance suitable for packed sequence attention kernels.

Parameters:

batch – A dictionary containing packed-sequence metadata. Expected keys: cu_seqlens, optional cu_seqlens_unpadded, optional argmins, and optional max_seqlen.

Returns:

PackedSeqParams with identical q/kv parameters and qkv_format set to “thd”.