Skip to content

vllm.model_executor.layers.mamba.ops.cpu.mamba_ssm

Functions:

selective_state_update(state, x, dt, A, B, C, D=None, z=None, dt_bias=None, dt_softplus=False, state_batch_indices=None, dst_state_batch_indices=None, null_block_id=NULL_BLOCK_ID, out=None, num_accepted_tokens=None, cu_seqlens=None, is_blackwell=False, enable_stochastic_rounding=False, cache_philox_rounds=0)

CPU implementation for selective_state_update.

Source code in vllm/model_executor/layers/mamba/ops/cpu/mamba_ssm.py 
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
def selective_state_update(
    state,
    x,
    dt,
    A,
    B,
    C,
    D=None,
    z=None,
    dt_bias=None,
    dt_softplus=False,
    state_batch_indices=None,
    dst_state_batch_indices=None,
    null_block_id=NULL_BLOCK_ID,
    out=None,
    num_accepted_tokens=None,
    cu_seqlens=None,
    is_blackwell=False,
    enable_stochastic_rounding=False,
    cache_philox_rounds=0,
):
    """CPU implementation for selective_state_update."""
    # Ensure out tensor exists
    if out is None:
        out = torch.empty_like(x if x.dim() == 2 else x)

    _state = state.unsqueeze(1) if state.dim() == 3 else state
    _x = x.unsqueeze(1) if x.dim() == 2 else x
    _dt = dt.unsqueeze(1) if dt.dim() == 2 else dt
    _A = A.unsqueeze(0) if A.dim() == 2 else A
    _B = B.unsqueeze(1) if B.dim() == 2 else B
    _C = C.unsqueeze(1) if C.dim() == 2 else C
    _D = D.unsqueeze(0) if (D is not None and D.dim() == 1) else D
    _z = z.unsqueeze(1) if (z is not None and z.dim() == 2) else z
    _dt_bias = (
        dt_bias.unsqueeze(0)
        if (dt_bias is not None and dt_bias.dim() == 1)
        else dt_bias
    )
    _out = out.unsqueeze(1) if out.dim() == 2 else out

    _sbi = state_batch_indices
    _dsbi = dst_state_batch_indices
    ops.selective_state_update_cpu(
        _state,
        _x,
        _dt,
        _A,
        _B,
        _C,
        _D,
        _z,
        _dt_bias,
        dt_softplus,
        _sbi,
        _dsbi,
        null_block_id,
        _out,
        num_accepted_tokens,
        cu_seqlens,
    )
    return _out.squeeze(1) if out.dim() == 2 else _out