From 9222d1a66b585d7ace851dcd710d5d76e220e191 Mon Sep 17 00:00:00 2001 From: slf12 Date: Wed, 30 Oct 2019 08:37:37 +0000 Subject: [PATCH] fix details --- paddleslim/quant/quant_embedding.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/paddleslim/quant/quant_embedding.py b/paddleslim/quant/quant_embedding.py index e5fdb120..8db9ddf7 100755 --- a/paddleslim/quant/quant_embedding.py +++ b/paddleslim/quant/quant_embedding.py @@ -170,7 +170,7 @@ def _quant_embedding_abs_max(graph, scope, place, config): (1 << (bit_length - 1)) - 1)) return scale, quanted_tensor.astype(np.int8) - def _insert_dequant_abx_max_op(graph, scope, var_node, scale_node, config): + def _insert_dequant_abs_max_op(graph, scope, var_node, scale_node, config): """ Insert dequantize_abs_max op in graph """ @@ -233,7 +233,7 @@ def _quant_embedding_abs_max(graph, scope, place, config): if op_node.name() == 'lookup_table': graph.update_input_link(embedding_node, quant_tensor_var, op_node) var_node = op_node.outputs[0] - _insert_dequant_abx_max_op(graph, scope, var_node, scale_var, + _insert_dequant_abs_max_op(graph, scope, var_node, scale_var, config) # free float embedding params memory -- GitLab