Skip to content
This repository was archived by the owner on Jul 7, 2023. It is now read-only.

Avoid error in beam search when "f" is in cache #1302

Merged
merged 4 commits into from
Jan 9, 2019
Merged

Avoid error in beam search when "f" is in cache #1302

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
189c5c9
add caching mechanism support for fast decoding with relative_dot_pro…
Dec 12, 2018
fa4fdee
fix typo
Dec 13, 2018
608be59
merge upstream
Dec 14, 2018
8b536bd
remove f in cache if use dense_relu_dense or conv_hidden_relu so that…
Dec 14, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 6 additions & 4 deletions tensor2tensor/models/transformer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -773,10 +773,11 @@ def fast_decode_tpu(encoder_output,
common_attention.split_heads(
tf.zeros([batch_size, decode_length, value_channels]),
hparams.num_heads),
"f":
tf.zeros([batch_size, decode_length, hparams.hidden_size]),
} for layer in range(num_layers)
}
if hparams.ffn_layer not in ["dense_relu_dense", "conv_hidden_relu"]:
for layer in range(num_layers):
cache["layer_%d" % layer]["f"] = tf.zeros([batch_size, 0, hparams.hidden_size])

if encoder_output is not None:
for layer in range(num_layers):
Expand Down Expand Up @@ -951,10 +952,11 @@ def fast_decode(encoder_output,
"v":
common_attention.split_heads(
tf.zeros([batch_size, 0, value_channels]), hparams.num_heads),
"f":
tf.zeros([batch_size, 0, hparams.hidden_size]),
} for layer in range(num_layers)
}
if hparams.ffn_layer not in ["dense_relu_dense", "conv_hidden_relu"]:
for layer in range(num_layers):
cache["layer_%d" % layer]["f"] = tf.zeros([batch_size, 0, hparams.hidden_size])

if encoder_output is not None:
for layer in range(num_layers):
Expand Down