cztomsik / ava

All-in-one desktop app for running LLMs locally.

Home Page:https://avapls.com

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Cannot download model

lming opened this issue · comments

Got Unexpected Error when clicking "Download" of a model, same error when clicking "Import" button.
Console shows error

01:17:31 debug default: Missing dependency: src.api.download.POST /download__struct_18238
01:17:31 debug server: POST /api/download 500

Both headless web app and mac app behaviors the same.
image

Hey, thanks for reporting, should be fixed now. I did some refactoring yesterday (extracting the server-code to its own repo) and looks like I messed up something, sorry. :)

Just wait for a new build if you are not building yourself.

Hey - thanks for the quick turn around - it still doesn't work for me

lming@mmax ava % zig build
Fetch Packages [3/3] tokamak... /Users/lming/work/ava/build.zig.zon:13:21: error: hash mismatch: expected: 12206b21956424fa8d78b5c959fac09247de583ace30e7440aebeba936e098d8fb3a, found: 12201735ec280c5444883ea131682f2eda70be65b120b661a7839dada3636f962959
            .hash = "12206b21956424fa8d78b5c959fac09247de583ace30e7440aebeba936e098d8fb3a",

If I modified the hash to the found value, it complains another error:

lming@mmax ava % zig build
warning: FileNotFound: /Users/lming/.cache/zig/p/1220491e02c1f645f41d46803d6fc039cbeef125a1d106f90a9cc718f6782df28a08/build.zig
error: FileNotFound

I am new to Zig so I don't know how to fix this error...

You best bet is to just grab a copy from github actions
https://github.com/cztomsik/ava/actions/runs/7774124705

Zig build system is a new thing, it might also have some bugs.

Cached files are in ~/.cache/zig on macos and probably something similar on linux. Also, always try with the newest zig version. I am updating every week at least.

EDIT: For macos builds, you also need ggml-metal.metal file to be in the same directory. You can find it in the llama.cpp repo. If you have time, you can try to figure out why the ./src/macos/create-dmg.sh does not work in the github pipeline, because that would the problem entirely and you could then just grab dmg there.

The original issue was fixed. Pipelines are also passing so I think either it works now or it has to be something specific to your machine.

I'd like to give this project a go, but I'm stuck on this as well. Grabbed a brand new headless build from pipeline but this is what I've got.
image

Can you check the Network tab in google chrome? I don't have linux machine here to check it out but macos headless build works fine (download starts and looks it goes as expected). Also, there is a tab with system logs in the Settings screen.

Sure, it's responding with
{"error":"MissingField"}
in response of POST: http://127.0.0.1:3002/api/models

here are logs from System tab:


16:17:02 debug server: GET /app.js 200
16:17:03 debug server: GET /api/chat 200
16:17:03 debug server: GET /api/models 200
16:17:07 debug server: GET / 200
16:17:07 debug server: GET /app.js 200
16:17:07 debug server: GET /api/models 200
16:17:07 debug server: GET /api/chat 200
16:17:14 debug server: GET /api/models 200
16:17:14 debug server: GET /api/system-info 200
16:18:50 debug server: POST /api/download 200
16:18:50 debug server: POST /api/models 500
16:18:50 debug server: GET /api/models 200

Apologies, there was one small bug but I think it will not build currently for linux because there's a problem in the llama.cpp repo. Here's what we are waiting for to get merged ggerganov/llama.cpp#5537