shithub: opus

Info  •  Files  •  Log  •  Branches

added .copy() to weights in wexchange

--- a/dnn/torch/weight-exchange/wexchange/torch/torch.py

+++ b/dnn/torch/weight-exchange/wexchange/torch/torch.py

@@ -39,14 +39,14 @@

     assert gru.num_layers == 1

     assert gru.bidirectional == False

-    w_ih = gru.weight_ih_l0.detach().cpu().numpy()

-    w_hh = gru.weight_hh_l0.detach().cpu().numpy()

+    w_ih = gru.weight_ih_l0.detach().cpu().numpy().copy()

+    w_hh = gru.weight_hh_l0.detach().cpu().numpy().copy()

     if hasattr(gru, 'bias_ih_l0'):

-        b_ih = gru.bias_ih_l0.detach().cpu().numpy()

+        b_ih = gru.bias_ih_l0.detach().cpu().numpy().copy()

     else:

         b_ih = None

     if hasattr(gru, 'bias_hh_l0'):

-        b_hh = gru.bias_hh_l0.detach().cpu().numpy()

+        b_hh = gru.bias_hh_l0.detach().cpu().numpy().copy()

     else:

         b_hh = None

@@ -81,11 +81,11 @@

 def dump_torch_dense_weights(where, dense, name='dense', scale=1/128, sparse=False, diagonal=False, quantize=False):

-    w = dense.weight.detach().cpu().numpy()

+    w = dense.weight.detach().cpu().numpy().copy()

     if dense.bias is None:

         b = np.zeros(dense.out_features, dtype=w.dtype)

     else:

-        b = dense.bias.detach().cpu().numpy()

+        b = dense.bias.detach().cpu().numpy().copy()

     if isinstance(where, CWriter):

         return print_dense_layer(where, name, w, b, scale=scale, format='torch', sparse=sparse, diagonal=diagonal, quantize=quantize)

@@ -110,11 +110,11 @@

 def dump_torch_conv1d_weights(where, conv, name='conv', scale=1/128, quantize=False):

-    w = conv.weight.detach().cpu().numpy()

+    w = conv.weight.detach().cpu().numpy().copy()

     if conv.bias is None:

         b = np.zeros(conv.out_channels, dtype=w.dtype)

     else:

-        b = conv.bias.detach().cpu().numpy()

+        b = conv.bias.detach().cpu().numpy().copy()

     if isinstance(where, CWriter):

@@ -141,7 +141,7 @@

 def dump_torch_embedding_weights(where, emb):

     os.makedirs(where, exist_ok=True)

-    w = emb.weight.detach().cpu().numpy()

+    w = emb.weight.detach().cpu().numpy().copy()

     np.save(os.path.join(where, 'weight.npy'), w)

-- 

⑨