Other LLM to create config files

E

escape

Does anyone know good pointers to read about creating configuration files using LLM? How about creating FreeBSD configuration files (or any application configuration files)?

Simple tests show that at least sometimes instructions given to the LLM how to create the config file should be improved. Is it possible to check if the config works and then instruct the LLM to re-create these config files after giving more instructions. The simple tests show that a feedback -system could help to improve the fine-tuning instructions given to the LLM and to ensure that the config works as expected. Are there any reinforcement learning cycles already tested, human monitored for example?

The LLM can be instructed by fine tuning and giving examples how to output data. For example config files.

esc
 
config files for what?
Base system FreeBSD default config is fine meaning you don't need to do anything.
I'd guess that 98% of FreeBSD users all they do is create a /etc/rc.conf to specify interfaces and what klds to load.
If you can read man pages, you don't need AI to load klds, enable zfs, set ethernet devices to DHCP.

It's not that difficult, I'd bet one would spend more time and energy "teaching" AI/LLB than it takes to manually create correct config by reading man pages or reading the Handbook.
 
Read, set, test, test again and forget. I mean enjoy.

I have used ollama to analyze my config files.

cat /etc/sysctl.conf | ollama run qwen3.5 --verbose "analyze this file please, give suggestions on improvements and present a new file with corrections if needed, comments are ignored - I might need the commented option for later use."
 
NeilAlieN75 said:
Read, set, test, test again and forget. I mean enjoy.

I have used ollama to analyze my config files.

cat /etc/sysctl.conf | ollama run qwen3.5 --verbose "analyze this file please, give suggestions on improvements and present a new file with corrections if needed, comments are ignored - I might need the commented option for later use."
Click to expand...
And what has been suggested? Is it only analyzing based on the contents of your sysctl.conf which is typically minimal or nonexistant? Most people go forever without any sysctl.conf at all.
Looking at min, I have some security thing, and changing the default tcp stack to bbr. So what suggestions could be offered?
 
collaboration of myself , internet & google ai :)
cat sysctl.conf
Code: 
# --- [ 1. SECURITY & JAILS ] ---
security.bsd.see_other_uids=1
security.bsd.see_other_gids=1
security.bsd.see_jail_proc=1
security.bsd.hardlink_check_uid=0
security.bsd.hardlink_check_gid=0
security.bsd.unprivileged_read_msgbuf=1
security.bsd.unprivileged_proc_debug=1
security.bsd.unprivileged_idprio=1
security.jail.sysvipc_allowed=1
security.jail.chflags_allowed=1
security.jail.allow_raw_sockets=1
kern.elf64.aslr.enable=1
kern.elf32.aslr.enable=1
kern.randompid=1

# --- [ 2. KERNEL & SYSTEM ] ---
kern.msgbuf_show_timestamp=1
kern.shutdown.kproc_shutdown_wait=20
kern.init_shutdown_timeout=180
kern.sched.preempt_thresh=85
kern.sched.slice=1

# --- [ 3. CORE DUMPS ] ---
kern.coredump=1
kern.sugid_coredump=1
kern.compress_user_cores=2
kern.corefile=/var/crash/%H.%N.%U.%P.%I.core

# --- [ 4. IPC & SHARED MEMORY ] ---
kern.ipc.shm_use_phys=1
kern.ipc.shmall=1048576
kern.ipc.shmmax=4294967296
kern.ipc.shm_allow_removed=1
kern.ipc.somaxconn=1280
kern.ipc.maxsockbuf=4194304

# --- [ 5. NETWORKING INTERN (IPv4 & LOCAL) ] ---
net.local.stream.recvspace=262144
net.local.stream.sendspace=262144
net.inet.tcp.sendspace=262144
net.inet.tcp.recvspace=262144
net.inet.tcp.cc.algorithm=cubic
net.inet.tcp.path_mtu_discovery=1
net.inet.tcp.pmtud_blackhole_detection=1
net.inet.tcp.pmtud_blackhole_mss=1536
net.inet.ip.random_id=1
net.inet.ip.forwarding=0
net.inet.ip.maxfragpackets=0
net.inet.ip.maxfragsperpacket=0

# --- [ 6. NETWORKING IPv6 ] ---
net.inet6.ip6.use_tempaddr=1
net.inet6.ip6.prefer_tempaddr=1
net.inet6.ip6.temppltime=7200
net.inet6.ip6.tempvltime=14400
net.inet6.ip6.accept_rtadv=1
net.inet6.ip6.fw.enable=1

# --- [ 7. VFS & STORAGE (ZFS) ] ---
kern.metadelay=4
kern.dirdelay=5
kern.filedelay=7
vfs.usermount=1
vfs.read_max=64
vfs.vmiodirenable=1
vfs.zfs.vdev.min_auto_ashift=12
vfs.zfs.arc.min=1500000000
vfs.zfs.arc.max=12000000000
vfs.zfs.txg.timeout=5

# --- [ 8. VM & MEMORY ] ---
vm.cluster_anon=1

# --- [ 9. HARDWARE & AUDIO ] ---
hw.usb.no_shutdown_wait=1
hw.usb.uaudio.default_rate=9600
hw.usb.uaudio.buffer_ms=1
hw.snd.feeder_rate_quality=3
hw.snd.latency=2
hw.snd.vpc_autoreset=0
hw.snd.default_unit=7
hw.usb.usbhid.enable=1
#dev.pcm.8.bitperfect=1
 
It's lets me know if a value is too large, whether it's a good setting, whether it's essential, obsolete, safe, etc. *I didn't need A.I. too tell me any of this to begin with because I did my research and testing my settings thoroughly beforehand. I just wanted to see what these models would suggest. My sysctl.conf is about 4.0k
 
NeilAlieN75 said:
It's lets me know if a value is too large, whether it's a good setting, whether it's essential, obsolete, safe, etc. *I didn't need A.I. too tell me any of this to begin with because I did my research and testing my settings thoroughly beforehand. I just wanted to see what these models would suggest. My sysctl.conf is about 4.0k
Click to expand...
That is bigger than anything I've run across. "Good setting" what determines good. Obsolete, sure that makes sense because some sysctl oids have changed across versions.
Value ranges: if they change across versions the LLM needs to know what version which your command does not seem to take into account
I'm not saying the process/tool (AI) is not useful, but rather the generic statements of the OP I find troublesome
 
You must log in or register to reply here.
Back
Top