llm-pi-zero/launch.sh at main · fxlin/llm-pi-zero · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
#!/bin/bash

# for pi-demo, configure the demo options
# 1. can be launched manually (must be as root),

# 2. by a systemd service (cf pi-demo.service)
# to install:
#       cp launch.sh /mnt/fat/launch.sh


# need to be in there to find font, etc
CODEDIR=/root/workspace-rwkv/llm-epaper-demo
# a fat partition to store the launch.sh and logs, so PC can read them
FATDIR=/mnt/fat

# RUN THE VANILLA RWKV INFERENCE ENGINE (x51, x52 only)
# installed w/ the vanilla rwkv
#source /root/workspace-rwkv/rootenv/bin/activate

# RUN OUR OWN RWKV INFERENCE ENGINE
# installed with our own rwkv packge
source /root/workspace-rwkv/myenv/bin/activate

####
# models
# official model
# export MODEL_PATH='/data/models/pi-deployment/RWKV-5-World-0.1B-v1-20230803-ctx4096'
# our own models (no cls, no mlp)
#export MODEL_PATH='/data/models/pi-deployment/01b-pre-x52-1455'
export MODEL_PATH='/data/models/pi-deployment/04b-pre-x59-2405'
# export MODEL_PATH='/data/models/pi-deployment/04b-tunefull-x58-562'
# export MODEL_PATH='/data/models/pi-deployment/1b5-pre-x59-929'
# export MODEL_PATH='/data/models/pi-deployment/1b5-pre-x59-929_fp16i8'     # works, but slow

# our own models (cls, mlp)
# export MODEL_PATH='/data/models/orin-deployment/01b-x59'
# export MODEL_PATH='/data/models/orin-deployment/04b-x59'

####
# prompts
export PROMPT_PATH='$FATDIR/prompts-qa.txt'
# export PROMPT_PATH='$FATDIR/prompts-topics.txt'

swapon /swapfile
swapoff /dev/zram0

cd $CODEDIR &&
python3 /root/workspace-rwkv/llm-epaper-demo/pi-demo.py  2>$FATDIR/pi-demo.err 1>$FATDIR/pi-demo.log