From fed30ef6294959080c29710555b4f401d85e994a Mon Sep 17 00:00:00 2001
From: slf12 <slf12thuss@163.com>
Date: Fri, 22 Nov 2019 14:42:16 +0800
Subject: [PATCH] save quanted model

---
 demo/quant/quant_embedding/README.md | 4 +++-
 demo/quant/quant_embedding/infer.py  | 4 ++++
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/demo/quant/quant_embedding/README.md b/demo/quant/quant_embedding/README.md
index 10d6f198..5667b19a 100755
--- a/demo/quant/quant_embedding/README.md
+++ b/demo/quant/quant_embedding/README.md
@@ -184,7 +184,7 @@ step:1 2722
 epoch:9          acc:0.153
 ```
 
-## 量化``基于skip-gram的word2vector模型``的训练流程保存的模型
+## 量化``基于skip-gram的word2vector模型``
 
 量化配置为:
 ```
@@ -236,3 +236,5 @@ quant_embedding config {'quantize_type': 'abs_max', 'params_name': 'emb', 'quant
 step:1 2719
 epoch:9          acc:0.153
 ```
+
+量化后的模型保存在``./output_quant``中，可看到量化后的参数``'emb.int8'``的大小为3.9M, 在``./v1_cpu5_b100_lr1dir``中可看到量化前的参数``'emb'``的大小为16M。
diff --git a/demo/quant/quant_embedding/infer.py b/demo/quant/quant_embedding/infer.py
index fa906bf9..40ae2ee8 100755
--- a/demo/quant/quant_embedding/infer.py
+++ b/demo/quant/quant_embedding/infer.py
@@ -82,6 +82,10 @@ def infer_epoch(args, vocab_size, test_reader, use_cuda, i2w):
                 if args.emb_quant:
                     config = {'params_name': 'emb', 'quantize_type': 'abs_max'}
                     copy_program = quant_embedding(copy_program, place, config)
+                    fluid.io.save_persistables(
+                        exe,
+                        './output_quant/pass-' + str(epoch),
+                        main_program=copy_program)
 
                 accum_num = 0
                 accum_num_sum = 0.0
-- 
GitLab