File size: 1,624 Bytes
d8bc908 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 | """Tests for LTI injection in ACT loops."""
import torch
from arbitor.components import LTIInjection, ByteHead
from arbitor.decoders import VideoHead
def test_lti_basic_properties():
lti = LTIInjection(64)
h = torch.randn(2, 10, 64)
e = torch.randn(2, 10, 64)
t = torch.randn(2, 10, 64)
out = lti(h, e, t)
assert out.shape == h.shape
assert torch.isfinite(out).all()
def test_lti_spectral_radius():
lti = LTIInjection(64)
A = lti.get_A()
assert (A > 0).all()
assert (A < 1).all()
def test_lti_learnable_params():
lti = LTIInjection(128)
assert lti.log_A.shape == (128,)
assert lti.log_dt.shape == (1,)
assert lti.B.shape == (128,)
assert sum(p.numel() for p in lti.parameters()) == 128 + 1 + 128
def test_lti_state_decay():
lti = LTIInjection(8)
h = torch.ones(1, 1, 8) * 100.0
e = torch.zeros(1, 1, 8)
t = torch.zeros(1, 1, 8)
out = lti(h, e, t)
assert (out.abs() < 50).all()
def test_lti_initial_state_small():
lti = LTIInjection(8)
h = torch.zeros(1, 1, 8)
e = torch.ones(1, 1, 8) * 5.0
t = torch.zeros(1, 1, 8)
out = lti(h, e, t)
assert (out > 0).all()
assert (out < 5).all()
def test_bytehead_lti_integration():
bh = ByteHead()
x = torch.randn(2, 10, 8192)
logits = bh(x)
assert logits.shape[-1] == 288
assert bh.lti is not None
assert isinstance(bh.lti, LTIInjection)
def test_bytehead_no_act():
bh_single = ByteHead(act_max_iters=1)
assert bh_single.lti is None
x = torch.randn(1, 5, 8192)
logits = bh_single(x)
assert logits.shape[-1] == 288
|