Chakshu Mira

Karma: 0

Chakshu Mira 2 May 2024 22:05 UTC
1 point
0
on: Ophiology (or, how the Mamba architecture works)
## Discretize B ## # [B,N] [E->N] [B,E] B = layer.W_B(x[b,l]) # no bias
Shouldn’t this be x[:,l] instead of x[b,l]?

Chakshu Mira 22 Apr 2024 20:46 UTC
1 point
0
in reply to: Adrià Garriga-alonso’s comment on: Ophiology (or, how the Mamba architecture works)
y_t=[N]C[E,N]h_t+[E] <this one> E [E]xt
Shouldn’t this be ‘D’?

Chakshu Mira 18 Apr 2024 21:42 UTC
1 point
0
on: Ophiology (or, how the Mamba architecture works)
E
Did you mean ‘D’ here? (2nd equation of the structured SSM)

Chakshu Mira 18 Apr 2024 1:08 UTC
1 point
0
on: Ophiology (or, how the Mamba architecture works)
Is this a typo? (Δtvt+1)xt−1