rm new unit tests, better covered by the integration test.

Author: alxmrs

tests/test_encoder.py

@@ -1,9 +1,8 @@
-import pytest
 import torch
 from perceiver_pytorch import Perceiver
 
 from ocean_emulators.constants import Lat, Lon
-from ocean_emulators.models.modules.encoder import PerceiverEncoder, patch_from
+from ocean_emulators.models.modules.encoder import PerceiverEncoder
 
 LATENT_DIM = 4
 
@@ -85,72 +84,3 @@ def test_makes_patches__more_variables():
     patches = encoder(prog, boundary, make_resolution(prog))
 
     assert patches.shape == (1, embed_dim, 1, 2)
-
-
-def test_cross_resolution_token_fuse():
-    """Prog at 1/4 degree (8x16) and boundary at 1 degree (2x4) produce the same latent grid."""
-    embed_dim = 4
-    prog = torch.randn(1, 7, 8, 16)
-    boundary = torch.randn(1, 3, 2, 4)
-
-    encoder = make_encoder(7, 3, embed_dim, (90, 90))
-
-    prog_res = (
-        torch.linspace(-90, 90, 8),
-        torch.linspace(0, 360, 16),
-    )
-    patches = encoder(prog, boundary, prog_res)
-
-    # Both grids with 90-degree patches → 2 lat patches, 4 lon patches
-    assert patches.shape == (1, embed_dim, 2, 4)
-    assert torch.isfinite(patches).all(), "Output contains NaN or Inf."
-
-
-def test_latent_grid_mismatch_raises():
-    """Misaligned latent grids between prog and boundary should raise."""
-
-    embed_dim = 4
-    # prog: 8x16 with patch_extent=(90,90) → ph=4,pw=4 → latent 2x4
-    # boundary: 4x6 with same extent → ph=2,pw=2 → latent 2x3 (mismatch on lon)
-    prog = torch.randn(1, 7, 8, 16)
-    boundary = torch.randn(1, 3, 4, 6)
-
-    encoder = make_encoder(7, 3, embed_dim, (90, 90))
-    prog_res = (torch.linspace(-90, 90, 8), torch.linspace(0, 360, 16))
-
-    with pytest.raises(AssertionError, match="Latent grid mismatch"):
-        encoder(prog, boundary, prog_res)
-
-
-def test_gradients_flow_to_both_streams():
-    """Gradients flow from the output back to both prognostic and boundary inputs."""
-    embed_dim = 4
-    prog = torch.randn(1, 7, 4, 8, requires_grad=True)
-    boundary = torch.randn(1, 3, 4, 8, requires_grad=True)
-
-    encoder = make_encoder(7, 3, embed_dim, (180, 180))
-    out = encoder(prog, boundary, make_resolution(prog))
-    out.sum().backward()
-
-    assert prog.grad is not None and prog.grad.abs().sum() > 0, (
-        "Gradients must flow to prognostic input."
-    )
-    assert boundary.grad is not None and boundary.grad.abs().sum() > 0, (
-        "Gradients must flow to boundary input."
-    )
-
-
-def test_patch_from__full_globe():
-    patch_h, patch_w = patch_from(
-        patch_extent=(180.0, 360.0), input_height=4, input_width=8
-    )
-    assert patch_h == 4
-    assert patch_w == 8
-
-
-def test_patch_from__half_extent():
-    patch_h, patch_w = patch_from(
-        patch_extent=(90.0, 180.0), input_height=4, input_width=8
-    )
-    assert patch_h == 2
-    assert patch_w == 4