local-ai: models are not getting downloaded

[mrwsl]

First of all: thanks for Nextcloud AIO. It is wonderful!

I tried to get the community container local-ai working, but it appears that it is not downloading any models from the models.yaml. The docker volume has a size of basically zero. When heading to the admin/ai settings page, I get a pop up that warns: "Failed to load models". The current file has:

- url: github:mudler/LocalAI/gallery/stablediffusion3.yaml
  name: Stable_diffusion

I tried it with the documented default as well. This is the log output of said container:

2025-01-02T13:17:30.739917200Z CPU info:
2025-01-02T13:17:30.741330120Z model name	: 12th Gen Intel(R) Core(TM) i9-12900K
2025-01-02T13:17:30.742075640Z flags		: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb intel_pt sha_ni xsaveopt xsavec xgetbv1 xsaves split_lock_detect avx_vnni dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp hwp_pkg_req hfi umip pku ospke waitpkg gfni vaes vpclmulqdq tme rdpid movdiri movdir64b fsrm md_clear serialize pconfig arch_lbr ibt flush_l1d arch_capabilities
2025-01-02T13:17:30.742618340Z CPU:    AVX    found OK
2025-01-02T13:17:30.743190988Z CPU:    AVX2   found OK
2025-01-02T13:17:30.743749645Z CPU: no AVX512 found
2025-01-02T13:17:30.746171600Z Waiting for nextcloud to start
2025-01-02T13:17:30.746180429Z nc: getaddrinfo for host "nextcloud-aio-nextcloud" port 9001: Name or service not known
2025-01-02T13:17:35.751213226Z nc: getaddrinfo for host "nextcloud-aio-nextcloud" port 9001: Name or service not known
2025-01-02T13:17:35.751536647Z Waiting for nextcloud to start
2025-01-02T13:17:40.754198859Z nc: getaddrinfo for host "nextcloud-aio-nextcloud" port 9001: Name or service not known
2025-01-02T13:17:40.754278092Z Waiting for nextcloud to start
2025-01-02T13:17:45.756054641Z Waiting for nextcloud to start
2025-01-02T13:17:50.757966268Z Waiting for nextcloud to start
2025-01-02T13:17:55.759831995Z Waiting for nextcloud to start
2025-01-02T13:18:00.761739315Z Waiting for nextcloud to start
2025-01-02T13:18:05.768106249Z ++ nproc
2025-01-02T13:18:05.769772124Z + THREADS=24
2025-01-02T13:18:05.769807694Z + export THREADS
2025-01-02T13:18:05.769819481Z + set +x
2025-01-02T13:18:05.802145390Z �[90m1:18PM�[0m �[32mINF�[0m �[1menv file found, loading environment variables from file�[0m �[36menvFile=�[0m.env
2025-01-02T13:18:05.803243495Z �[90m1:18PM�[0m �[32mINF�[0m �[1mSetting logging to info�[0m
2025-01-02T13:18:05.803252464Z �[90m1:18PM�[0m �[32mINF�[0m �[1mStarting LocalAI using 24 threads, with models path: /models�[0m
2025-01-02T13:18:05.803255087Z �[90m1:18PM�[0m �[32mINF�[0m �[1mLocalAI version: v2.24.2 (59cf30a80e40236b9ff4b36e527fadff2f3fc0c3)�[0m
2025-01-02T13:18:05.852121534Z �[90m1:18PM�[0m �[32mINF�[0m �[1mPreloading models from /models�[0m
2025-01-02T13:18:06.152243952Z �[90m1:18PM�[0m �[32mINF�[0m �[1mcore/startup process completed!�[0m
2025-01-02T13:18:06.152969084Z �[90m1:18PM�[0m �[32mINF�[0m �[1mLocalAI API is listening! Please connect to the endpoint for API documentation.�[0m �[36mendpoint=�[0mhttp://0.0.0.0:8080
2025-01-02T13:18:30.824931887Z �[90m1:18PM�[0m �[32mINF�[0m �[1mSuccess�[0m �[36mip=�[0m127.0.0.1 �[36mlatency=�[0m"29.37µs" �[36mmethod=�[0mGET �[36mstatus=�[0m200 �[36murl=�[0m/readyz
2025-01-02T13:19:30.912247972Z �[90m1:19PM�[0m �[32mINF�[0m �[1mSuccess�[0m �[36mip=�[0m127.0.0.1 �[36mlatency=�[0m"20.186µs" �[36mmethod=�[0mGET �[36mstatus=�[0m200 �[36murl=�[0m/readyz
2025-01-02T13:20:31.014921065Z �[90m1:20PM�[0m �[32mINF�[0m �[1mSuccess�[0m �[36mip=�[0m127.0.0.1 �[36mlatency=�[0m"29.833µs" �[36mmethod=�[0mGET �[36mstatus=�[0m200 �[36murl=�[0m/readyz
2025-01-02T13:21:31.058090845Z �[90m1:21PM�[0m �[32mINF�[0m �[1mSuccess�[0m �[36mip=�[0m127.0.0.1 �[36mlatency=�[0m"18.64µs" �[36mmethod=�[0mGET �[36mstatus=�[0m200 �[36murl=�[0m/readyz

Any idea what I have missed?

EDIT: In the docker volumes is a yaml file:

$cat /var/lib/docker/volumes/nextcloud_aio_localai_models/_data/._gallery_Stable_diffusion.yaml
description: ""
icon: ""
license: ""
urls: []
name: ""
config_file: ""
files: []
prompt_templates: []

[szaimen]

Hi, can you please check if the following works? See d4a3bbe

[mrwsl]

I tried your approach unsuccessfully. I also tried the url - url: github:mudler/LocalAI/gallery/stablediffusion3.yaml@master name: Stable_diffusion like it is shown in the gallery document but the model isn't downloaded. At least another file got created:

drwxr-xr-x 2 root root 133 Jan  3 19:48 .
drwx-----x 3 root root  19 Dec  3 13:00 ..
-rw------- 1 root root 102 Dec  4 10:00 ._gallery_llama3.1.yaml
-rw------- 1 root root 102 Jan  1 05:03 ._gallery_Llama.yaml
-rw------- 1 root root 379 Jan  4 05:03 ._gallery_Stable_diffusion.yaml
-rw------- 1 root root 248 Jan  4 05:03 Stable_diffusion.yaml

[mrwsl]

To add: The documentation is using the raw.* urls. But I am not sure if --preload-models is taking care of downloading the models? It appears that it is only loading the config.

[roelofz]

Hi,
I have been tinkering with this this weekend and I also can not get the example to work or find a working url.
I did get one yaml in the models folder working, but this model did not seem to work properly.
So it seems very difficult toi add a working url from the mudler github repo.
It could help if we had a few models as an example, one for text, one for audio and one for image generation (maybe green ones, low in cpu and ethically responsible following NC's policies).
Downside of this is maintenance of the url's.

Maybe we can prevent that by including the examples in the own repo, with mentioning they are examples.
If the url model is working, a new user may be able to add own urls's, learn from it.