mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-08-26 18:24:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix test failures for torch 1.8.0

Browse Source
This commit is contained in:
Fangjun Kuang 2021-12-06 10:42:01 +08:00
parent b3a5b04e13
commit cafd06e909
1 changed files with 36 additions and 41 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

77
egs/librispeech/ASR/transducer/test_rnn.py
View File

@ -34,8 +34,10 @@ def get_devices():
return devices return devices
def assert_allclose(a: torch.Tensor, b: torch.Tensor, **kwargs): def assert_allclose(a: torch.Tensor, b: torch.Tensor, atol=1e-6, **kwargs):
assert torch.allclose(a, b, **kwargs), f"{(a - b).abs().max()}, {a.numel()}" assert torch.allclose(
a, b, atol=atol, **kwargs
), f"{(a - b).abs().max()}, {a.numel()}"
def test_layernorm_lstm_cell_jit(device="cpu"): def test_layernorm_lstm_cell_jit(device="cpu"):
@ -65,8 +67,7 @@ def test_layernorm_lstm_cell_constructor(device="cpu"):
torch_cell = nn.LSTMCell( torch_cell = nn.LSTMCell(
input_size, input_size,
hidden_size, hidden_size,
device=device, ).to(device)
)
for name, param in self_cell.named_parameters(): for name, param in self_cell.named_parameters():
assert param.shape == getattr(torch_cell, name).shape assert param.shape == getattr(torch_cell, name).shape
@ -103,8 +104,7 @@ def test_layernorm_lstm_cell_forward(device="cpu"):
input_size, input_size,
hidden_size, hidden_size,
bias=bias, bias=bias,
device=device, ).to(device)
)
with torch.no_grad(): with torch.no_grad():
for name, torch_param in torch_cell.named_parameters(): for name, torch_param in torch_cell.named_parameters():
self_param = getattr(self_cell, name) self_param = getattr(self_cell, name)
@ -121,7 +121,7 @@ def test_layernorm_lstm_cell_forward(device="cpu"):
torch_h, torch_c = torch_cell(x_clone, (h, c)) torch_h, torch_c = torch_cell(x_clone, (h, c))
assert_allclose(self_h, torch_h) assert_allclose(self_h, torch_h)
assert_allclose(self_c, torch_c, atol=1e-6) assert_allclose(self_c, torch_c)
self_hc = self_h * self_c self_hc = self_h * self_c
torch_hc = torch_h * torch_c torch_hc = torch_h * torch_c
@ -155,8 +155,7 @@ def test_layernorm_lstm_cell_with_projection_forward(device="cpu"):
bias=bias, bias=bias,
proj_size=proj_size, proj_size=proj_size,
batch_first=True, batch_first=True,
device=device, ).to(device)
)
with torch.no_grad(): with torch.no_grad():
for name, self_param in self_cell.named_parameters(): for name, self_param in self_cell.named_parameters():
getattr(torch_cell, f"{name}_l0").copy_(self_param) getattr(torch_cell, f"{name}_l0").copy_(self_param)
@ -177,7 +176,7 @@ def test_layernorm_lstm_cell_with_projection_forward(device="cpu"):
torch_c = torch_c.squeeze(0) torch_c = torch_c.squeeze(0)
assert_allclose(self_h, torch_h) assert_allclose(self_h, torch_h)
assert_allclose(self_c, torch_c, atol=1e-6) assert_allclose(self_c, torch_c)
(self_h.sum() * self_c.sum()).backward() (self_h.sum() * self_c.sum()).backward()
(torch_h.sum() * torch_c.sum()).backward() (torch_h.sum() * torch_c.sum()).backward()
@ -244,8 +243,7 @@ def test_layernorm_lstm_layer_with_projection_forward(device="cpu"):
batch_first=True, batch_first=True,
dropout=0, dropout=0,
bidirectional=False, bidirectional=False,
device=device, ).to(device)
)
with torch.no_grad(): with torch.no_grad():
for name, self_param in self_layer.cell.named_parameters(): for name, self_param in self_layer.cell.named_parameters():
getattr(torch_layer, f"{name}_l0").copy_(self_param) getattr(torch_layer, f"{name}_l0").copy_(self_param)
@ -253,9 +251,9 @@ def test_layernorm_lstm_layer_with_projection_forward(device="cpu"):
torch_y, (torch_h, torch_c) = torch_layer( torch_y, (torch_h, torch_c) = torch_layer(
x_clone, (h.unsqueeze(0), c.unsqueeze(0)) x_clone, (h.unsqueeze(0), c.unsqueeze(0))
) )
assert_allclose(self_y, torch_y, atol=1e-6) assert_allclose(self_y, torch_y)
assert_allclose(self_h, torch_h, atol=1e-6) assert_allclose(self_h, torch_h)
assert_allclose(self_c, torch_c, atol=1e-6) assert_allclose(self_c, torch_c)
self_y.sum().backward() self_y.sum().backward()
torch_y.sum().backward() torch_y.sum().backward()
@ -294,8 +292,7 @@ def test_layernorm_lstm_layer_forward(device="cpu"):
batch_first=True, batch_first=True,
dropout=0, dropout=0,
bidirectional=False, bidirectional=False,
device=device, ).to(device)
)
with torch.no_grad(): with torch.no_grad():
for name, self_param in self_layer.cell.named_parameters(): for name, self_param in self_layer.cell.named_parameters():
getattr(torch_layer, f"{name}_l0").copy_(self_param) getattr(torch_layer, f"{name}_l0").copy_(self_param)
@ -303,9 +300,9 @@ def test_layernorm_lstm_layer_forward(device="cpu"):
torch_y, (torch_h, torch_c) = torch_layer( torch_y, (torch_h, torch_c) = torch_layer(
x_clone, (h.unsqueeze(0), c.unsqueeze(0)) x_clone, (h.unsqueeze(0), c.unsqueeze(0))
) )
assert_allclose(self_y, torch_y, atol=1e-6) assert_allclose(self_y, torch_y)
assert_allclose(self_h, torch_h, atol=1e-6) assert_allclose(self_h, torch_h)
assert_allclose(self_c, torch_c, atol=1e-6) assert_allclose(self_c, torch_c)
self_hc = self_h * self_c self_hc = self_h * self_c
torch_hc = torch_h * torch_c torch_hc = torch_h * torch_c
@ -386,8 +383,7 @@ def test_layernorm_lstm_forward(device="cpu"):
bias=bias, bias=bias,
batch_first=True, batch_first=True,
bidirectional=False, bidirectional=False,
device=device, ).to(device)
)
assert len(self_lstm.state_dict()) == len(torch_lstm.state_dict()) assert len(self_lstm.state_dict()) == len(torch_lstm.state_dict())
with torch.no_grad(): with torch.no_grad():
for name, param in self_lstm.named_parameters(): for name, param in self_lstm.named_parameters():
@ -457,8 +453,7 @@ def test_layernorm_lstm_with_projection_forward(device="cpu"):
proj_size=proj_size, proj_size=proj_size,
batch_first=True, batch_first=True,
bidirectional=False, bidirectional=False,
device=device, ).to(device)
)
assert len(self_lstm.state_dict()) == len(torch_lstm.state_dict()) assert len(self_lstm.state_dict()) == len(torch_lstm.state_dict())
with torch.no_grad(): with torch.no_grad():
for name, param in self_lstm.named_parameters(): for name, param in self_lstm.named_parameters():
@ -483,13 +478,13 @@ def test_layernorm_lstm_with_projection_forward(device="cpu"):
c = torch.stack(cs) c = torch.stack(cs)
torch_y, (torch_h, torch_c) = torch_lstm(x_clone, (h, c)) torch_y, (torch_h, torch_c) = torch_lstm(x_clone, (h, c))
assert_allclose(self_y, torch_y, atol=1e-6) assert_allclose(self_y, torch_y)
self_h = torch.stack([s[0] for s in self_states]) self_h = torch.stack([s[0] for s in self_states])
self_c = torch.stack([s[1] for s in self_states]) self_c = torch.stack([s[1] for s in self_states])
assert_allclose(self_h, torch_h, atol=1e-6) assert_allclose(self_h, torch_h)
assert_allclose(self_c, torch_c, atol=1e-6) assert_allclose(self_c, torch_c)
s = self_y.reshape(-1) s = self_y.reshape(-1)
t = torch_y.reshape(-1) t = torch_y.reshape(-1)
@ -501,7 +496,7 @@ def test_layernorm_lstm_with_projection_forward(device="cpu"):
shc_sum.backward() shc_sum.backward()
thc_sum.backward() thc_sum.backward()
assert_allclose(x.grad, x_clone.grad, atol=1e-6) assert_allclose(x.grad, x_clone.grad)
def test_layernorm_gru_cell_jit(device="cpu"): def test_layernorm_gru_cell_jit(device="cpu"):
@ -530,8 +525,7 @@ def test_layernorm_gru_cell_constructor(device="cpu"):
torch_cell = nn.GRUCell( torch_cell = nn.GRUCell(
input_size, input_size,
hidden_size, hidden_size,
device=device, ).to(device)
)
for name, param in self_cell.named_parameters(): for name, param in self_cell.named_parameters():
assert param.shape == getattr(torch_cell, name).shape assert param.shape == getattr(torch_cell, name).shape
@ -555,8 +549,7 @@ def test_layernorm_gru_cell_forward(device="cpu"):
input_size, input_size,
hidden_size, hidden_size,
bias=bias, bias=bias,
device=device, ).to(device)
)
with torch.no_grad(): with torch.no_grad():
for name, torch_param in torch_cell.named_parameters(): for name, torch_param in torch_cell.named_parameters():
self_param = getattr(self_cell, name) self_param = getattr(self_cell, name)
@ -580,7 +573,7 @@ def test_layernorm_gru_cell_forward(device="cpu"):
torch_h.reshape(-1) * torch.arange(torch_h.numel(), device=device) torch_h.reshape(-1) * torch.arange(torch_h.numel(), device=device)
).sum().backward() ).sum().backward()
assert_allclose(x.grad, x_clone.grad, atol=1e-4) assert_allclose(x.grad, x_clone.grad, atol=1e-3)
def test_layernorm_gru_layer_jit(device="cpu"): def test_layernorm_gru_layer_jit(device="cpu"):
@ -624,15 +617,14 @@ def test_layernorm_gru_layer_forward(device="cpu"):
batch_first=True, batch_first=True,
dropout=0, dropout=0,
bidirectional=False, bidirectional=False,
device=device, ).to(device)
)
with torch.no_grad(): with torch.no_grad():
for name, self_param in self_layer.cell.named_parameters(): for name, self_param in self_layer.cell.named_parameters():
getattr(torch_layer, f"{name}_l0").copy_(self_param) getattr(torch_layer, f"{name}_l0").copy_(self_param)
torch_y, torch_h = torch_layer(x_clone, h.unsqueeze(0)) torch_y, torch_h = torch_layer(x_clone, h.unsqueeze(0))
assert_allclose(self_y, torch_y, atol=1e-6) assert_allclose(self_y, torch_y)
assert_allclose(self_h, torch_h, atol=1e-6) assert_allclose(self_h, torch_h)
self_y_sum = ( self_y_sum = (
self_y.reshape(-1) * torch.arange(self_y.numel(), device=device) self_y.reshape(-1) * torch.arange(self_y.numel(), device=device)
@ -685,8 +677,7 @@ def test_layernorm_gru_forward(device="cpu"):
bias=bias, bias=bias,
batch_first=True, batch_first=True,
bidirectional=False, bidirectional=False,
device=device, ).to(device)
)
assert len(self_gru.state_dict()) == len(torch_gru.state_dict()) assert len(self_gru.state_dict()) == len(torch_gru.state_dict())
with torch.no_grad(): with torch.no_grad():
for name, param in self_gru.named_parameters(): for name, param in self_gru.named_parameters():
@ -709,11 +700,11 @@ def test_layernorm_gru_forward(device="cpu"):
torch_y, torch_states = torch_gru(x_clone, torch.stack(states)) torch_y, torch_states = torch_gru(x_clone, torch.stack(states))
assert_allclose(self_y, torch_y, atol=1e-6) assert_allclose(self_y, torch_y)
self_states = torch.stack(self_states) self_states = torch.stack(self_states)
assert_allclose(self_states, torch_states, atol=1e-6) assert_allclose(self_states, torch_states)
s = self_y.reshape(-1) s = self_y.reshape(-1)
t = torch_y.reshape(-1) t = torch_y.reshape(-1)
@ -758,6 +749,10 @@ def _test_gru(device):
test_layernorm_gru_forward(device) test_layernorm_gru_forward(device)
torch.set_num_threads(1)
torch.set_num_interop_threads(1)
def main(): def main():
for device in get_devices(): for device in get_devices():
print("device", device) print("device", device)