dictionary from unstructure log data can not speed up compress processing?

zstd version:
*** Zstandard CLI (32-bit) v1.5.4, by Yann Collet ***

My steps:
1. train from the original log file (20MB/file, 45 log files)
```
$ zstd -B4096 --train -r train
Trying 5 different sets of parameters
k=1024
d=8
f=20
steps=4
split=75
accel=1
Save dictionary of size 112640 into file dictionary
```

2. compress the original log file(used by trainning) and add contrast group without dictionary
```
time /tmp/zstd -k -1 -f -D dictionary -r train
45 files compressed : 9.28% ( 897 MiB => 83.3 MiB) ==> 9%
real 1m 12.22s
user 0m 30.25s
sys 0m 6.19s

time /tmp/zstd -k -1 -f -r train
45 files compressed : 9.27% ( 897 MiB => 83.2 MiB) ==> 9%
real 1m 5.28s
user 0m 29.67s
sys 0m 6.18s
```

My trainning dictionary has no effect on compress ratio and compress speed.
Thanks for your help!

In addition:
1. I train the dictionary in my macbookpro M1 (16 GB memory)
2. The compress testing run in linux with poor performance(armv7, 32-bit, 750 MB memory)
3. The log content like this(as the following screenshot)