Anybody else having problems with AutoTsgger?

[jamiesaun85]

Before the last update to Version 2025.2.4, Autotagger has had several issues. LLava 7 still works the best, but even it is now doing the same thing. Autotagger just sops working, and how many it will do seems totally random. If I cue up, say 100 files to annalyze and tag, it mwy do anywhere from 5 to 75 or whatever, but eventually it will stick on one and not move from there.

Also, this is with LLava 7b. If I use any other, including LLava 13, it will only do 3 or 4 befoore it stops. Gemma is particullary bad. It will only do 1 to 2 files at  a time and thats it. It will stop working after that and i have to clear the cue and try again.

Another note, when using the best one, LLava 7, it only works better if I dont touch it. I cant scroll and see what its filling out. I cant switch to another page and work on other photos and I cant switch to another app and and do something else on the pc  while its working. If I dont touch it, it will do a couple hundred or so before it chkes up and stops. If i touch it, it will basically stop immediatley and noot work again until I clear cue and run it again.

Before the update, LLava 7 would run an entire cue without ever stopping. I switched to a studio driver but it made no difference.

Is it just me or does anybody else have this problem?

[Mario]

Include a log file in debug logging mode (log file) to provide a minimum of information to work with.

Which type of hardware? Graphic card? VRAM?
Which AI runner do you use? Ollama or LM Studio?
Also ZIP and attach the log files of Ollama / LM Studio.

[jamiesaun85]

Will do. I use an RTX 4070ti super with 16GB VRAM. An i9 14900k and 64 gigs of RAM. I use Ollama, I tried LM studio but it was even worse. Ollama still works alright, but only with llAVA 7b and not nearly as well as it did before the update. Ill get some log files uploaded.

[Mario]

I use Ollama, I tried LM studio but it was even worse.

This indicates some issue with your computer or the installed graphic card drivers / CUDA drivers.
If both GPU-intensive software produces cause problems, it's not IMatch.

I run both Ollama and LM Studio on a 4070TI myself, every day. I've processed over 5,000 images over the weekend, without any problems. Using the Gemma 3 12B model. Which is way better than LLaVA or LLama Vision.

One user reported constant crashes with a specific NVIDIA driver version. Installing an older version solved the problem.

Note that both Ollama and LM Studio are updated once a week or more often. Make sure you install all available updates.
They often contain fixes for specific graphic cards / driver versions.

[jamiesaun85]

Okay, I rolled back to a driver from December, and it didn't help. Then I went back to the latest studio driver and it's worse now than ever. It won't run more than 1 tag and it just stops. The log mentions something about not connecting to the server, but the server is local, isn't it?

Thanks in advance.

[Mario]

18:08:00 Ollama starts
18:09:56 Ollama shuts down.

The log file is "C:\\Users\\jamie\\AppData\\Local\\Ollama\\server.log"

I see two successful requests in the log file. No warning, no error.

[jamiesaun85]

So the two successful requests must be two that it tags, and then it stops, But I never shut it down? I'm not sure what it means by that. I did just run it long enough to capture the issue, but it couldn't have been shut down because I Right-clicked on ollama to get the log file. There's lots more that are longer, i didn't think you would want a big long one. But ill grab a few more and post them.

By shut down, does it mean I cancelled the cue? Because that could definitely be it.

[jamiesaun85]

Heres some that are server logs instead of app logs

[Mario]

IMatch cannot shut down Ollama. Windows can. Ollama can shut down itself, probably?

I'm no Ollama expert. But looking at the log I see lines like

loaded CUDA backend from C:\Users\jamie\AppData\Local\Programs\Ollama\lib\ollama\cuda_v12\ggml-cuda.dll
level=ERROR source=ggml.go:88 msg="failed to get absolute path" error="The filename, directory name, or volume label syntax is incorrect."

I see no such error messages in my Ollama log files.

CUDA is software from NVIDIA which is required to allow software to use NVIDIA cards for AI.
AFAIK Ollama includes the CUDA libraries it needs and installs them automatically.
It seems to me that something is wrong and Ollama cannot load one or more DLLs it needs to run?

Maybe the Ollama installation is damaged? Try to uninstall and re-install it via the Windows Add/Remove software applet.

[jamiesaun85]

Yes sir, I will do that. I will also uninstall LM Studio and wipe all models and start fresh.

Thank you.

[Mario]

I'm currently preferring LM Studio, because it also allows me to chat with the loaded model. Especially useful with the powerful Gemma 3 12B model.

Ollama is simpler to use, though.

[Stenis]

Before the last update to Version 2025.2.4, Autotagger has had several issues. LLava 7 still works the best, but even it is now doing the same thing. Autotagger just sops working, and how many it will do seems totally random. If I cue up, say 100 files to annalyze and tag, it mwy do anywhere from 5 to 75 or whatever, but eventually it will stick on one and not move from there.

Also, this is with LLava 7b. If I use any other, including LLava 13, it will only do 3 or 4 befoore it stops. Gemma is particullary bad. It will only do 1 to 2 files at  a time and thats it. It will stop working after that and i have to clear the cue and try again.

Another note, when using the best one, LLava 7, it only works better if I dont touch it. I cant scroll and see what its filling out. I cant switch to another page and work on other photos and I cant switch to another app and and do something else on the pc  while its working. If I dont touch it, it will do a couple hundred or so before it chkes up and stops. If i touch it, it will basically stop immediatley and noot work again until I clear cue and run it again.

Before the update, LLava 7 would run an entire cue without ever stopping. I switched to a studio driver but it made no difference.

Is it just me or does anybody else have this problem?

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

[Mario]

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

Under which condition does AutoTagger (hang)?
What do you do, exactly, to produce this situation?
Which runner do you use, Ollama or LM Studio? Or do you use cloud-AI?
Please always keep the log file (Help menu > Copy Log File) and attach ZIPped when you run into such a problem.

I have worked with AutoTagger since Saturday many hours each day, processing over 10K images.
Using different settings, run once and run for all etc.
I've used LM Studio with Gemma3, OpenAI, and Gemini.
AutoTagger did not hang once.
I had to restart LM studio once because it no longer accepted requests and AutoTagger ran into a timeout.
I had to wait for a couple of minutes since response times from OpenAI went trough the roof.
That's it. Pretty stable.

[Stenis]

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

Under which condition does AutoTagger (hang)?
What do you do, exactly, to produce this situation?
Which runner do you use, Ollama or LM Studio? Or do you use cloud-AI?
Please always keep the log file (Help menu > Copy Log File) and attach ZIPped when you run into such a problem.

I have worked with AutoTagger since Saturday many hours each day, processing over 10K images.
Using different settings, run once and run for all etc.
I've used LM Studio with Gemma3, OpenAI, and Gemini.
AutoTagger did not hang once.
I had to restart LM studio once because it no longer accepted requests and AutoTagger ran into a timeout.
I had to wait for a couple of minutes since response times from OpenAI went trough the roof.
That's it. Pretty stable.

I have notised that sometimes when I have some pending updates and selected some pictures - Pressed F7 - and then "Run" Autotagger it might not start at all updating despite the update-symbol for Autotagger" is to be seen on the thumbnails. If I will force it to start it always seems to get released from that hanging and starts to process the Autotagger marked pictures. I definitely have verified that behavior with Gemma since the processing with Gemma imposes activity on my GPU and that sounds. Otherwise me Gemma 3 4 is pretty fast and finishes a picture in 3-5 seconds.

[jamiesaun85]

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

Under which condition does AutoTagger (hang)?
What do you do, exactly, to produce this situation?
Which runner do you use, Ollama or LM Studio? Or do you use cloud-AI?
Please always keep the log file (Help menu > Copy Log File) and attach ZIPped when you run into such a problem.

I have worked with AutoTagger since Saturday many hours each day, processing over 10K images.
Using different settings, run once and run for all etc.
I've used LM Studio with Gemma3, OpenAI, and Gemini.
AutoTagger did not hang once.
I had to restart LM studio once because it no longer accepted requests and AutoTagger ran into a timeout.
I had to wait for a couple of minutes since response times from OpenAI went trough the roof.
That's it. Pretty stable.

I have notised that sometimes when I have some pending updates and selected some pictures - Pressed F7 - and then "Run" Autotagger it might not start at all updating despite the update-symbol for Autotagger" is to be seen on the thumbnails. If I will force it to start it always seems to get released from that hanging and starts to process the Autotagger marked pictures. I definitely have verified that behavior with Gemma since the processing with Gemma imposes activity on my GPU and that sounds. Otherwise me Gemma 3 4 is pretty fast and finishes a picture in 3-5 seconds.

I see this too sometimes. The little Icon appears on the thumbnails, but nothing happens. Always when i hit alt-f7. When its hanging\, if I then hit f7 and then run, it will then double the cue and start running. But it never makes it very far.

Main issue is it only works for a oicture or two. Then nothing. It just stops. Im right now trying different drivers, LM studio, Ollama and all different models to see which may finaly fix something.

[jamiesaun85]

Hang on a tick. If It's ai, maybe i should ask it why it's not working correctly? Seems reasonable. Maybe it will know. LOL

I've never talked to AI before.

[jamiesaun85]

Well, that was very interesting, and a little unsettling, to be honest. But also very helpful. She, it? Gave me a few things to try within the configuration settings of ollama and LM Studio. I will try those and report back

[jamiesaun85]

Before the last update to Version 2025.2.4, Autotagger has had several issues. LLava 7 still works the best, but even it is now doing the same thing. Autotagger just sops working, and how many it will do seems totally random. If I cue up, say 100 files to annalyze and tag, it mwy do anywhere from 5 to 75 or whatever, but eventually it will stick on one and not move from there.

Also, this is with LLava 7b. If I use any other, including LLava 13, it will only do 3 or 4 befoore it stops. Gemma is particullary bad. It will only do 1 to 2 files at  a time and thats it. It will stop working after that and i have to clear the cue and try again.

Another note, when using the best one, LLava 7, it only works better if I dont touch it. I cant scroll and see what its filling out. I cant switch to another page and work on other photos and I cant switch to another app and and do something else on the pc  while its working. If I dont touch it, it will do a couple hundred or so before it chkes up and stops. If i touch it, it will basically stop immediatley and noot work again until I clear cue and run it again.

Before the update, LLava 7 would run an entire cue without ever stopping. I switched to a studio driver but it made no difference.

Is it just me or does anybody else have this problem?

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

I will try this also. Thank you

[jamiesaun85]

Before the last update to Version 2025.2.4, Autotagger has had several issues. LLava 7 still works the best, but even it is now doing the same thing. Autotagger just sops working, and how many it will do seems totally random. If I cue up, say 100 files to annalyze and tag, it mwy do anywhere from 5 to 75 or whatever, but eventually it will stick on one and not move from there.

Also, this is with LLava 7b. If I use any other, including LLava 13, it will only do 3 or 4 befoore it stops. Gemma is particullary bad. It will only do 1 to 2 files at  a time and thats it. It will stop working after that and i have to clear the cue and try again.

Another note, when using the best one, LLava 7, it only works better if I dont touch it. I cant scroll and see what its filling out. I cant switch to another page and work on other photos and I cant switch to another app and and do something else on the pc  while its working. If I dont touch it, it will do a couple hundred or so before it chkes up and stops. If i touch it, it will basically stop immediatley and noot work again until I clear cue and run it again.

Before the update, LLava 7 would run an entire cue without ever stopping. I switched to a studio driver but it made no difference.

Is it just me or does anybody else have this problem?

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

Well that was fast. Dude, that totally worked. You are so awesome! Thank you.

Mario, It hangs, or stops writing anything. I press alt+shift+S and it writes back the pending metadata. I touch nothing else and autotagger starts working and finished the cue. Not sure if this information helps you or not. But Stenis is very much correct. That worked like a charm.

[jamiesaun85]

Okay. I cue up 287 files to tag. Autotagger hangs after the very first one. After pushing alt, shift, S, it writes back the data. Autotagger then starts working and finishes the entire 286 left in the cue without any issue at all.

What's more, I can now go do other things within IMatch, and use other software entirely, and Autotaggewr keeps chugging along without issue. For whatever reason, this write back trick completely fixes the issue. After it's done with this latest cue, I will zip the log and load it here.

[jamiesaun85]

Here it is. Hopefully this helps. I am not a software guy, Im a hardware guy and an elctrical guy.

[Mario]

When you say "hang", I think you mean that the files you have enqueued for AutoTagger processing are added to the background processing queue but IMatch does not process them. When you look at the Info & Activity panel, what doe you see? Are there entries reported in the queue? Do any progress indicators show?

It seems that when the queue is stuck for some unknown reason on your PC, triggering a write-back, which pauses and un-pauses background processing, gets the queue running again.

Looking at your log file, I see this:

AutoTagger: Aborting because of error [530] 'PTCAIConnectorLMStudio: HTTP error.'

Error 530 means "HTTP error". IMatch uses HTTP to communicate with Ollama and LM Studio. Both offer their services by starting a web server. Apparently,. 530 is not a timeout, but something "else". LM Studio did not respond in a supported way, but also there was no status message with additional information. Else IMatch would have logged it.

AutoTagger cannot community with LM Studio or gets unexpected responses.

It automatically pauses for one minute or so and then tries again. it is not uncommon for cloud-based services like OpenAI to be temporarily unavailable due to overload, and AutoTagger deals with this by waiting a while and then automatically retrying.

I have never seen this with Ollama or LM Studio. Since this seems to affect only you, do you run any specific software? Which virus checker are you using? Did it log anything? Virus checkers might consider LM Studio a treat, because it opens a port and receives "commands"...

Do you see the "Files with problems" AutoTagger notification at the bottom of the IMatch window

[jamiesaun85]

When you say "hang", I think you mean that the files you have enqueued for AutoTagger processing are added to the background processing queue but IMatch does not process them. When you look at the Info & Activity panel, what doe you see? Are there entries reported in the queue? Do any progress indicators show?

It shows AItagger as normal. a red box that has a loading bar and a counter. Say 51/350. but it isnt going any higher than 51. Bottom right corner still shows AItagger as being busy doing something. But the blue circle thing is no longer spinning and no new files are flashing across the bottom. And the time out counter just gets longer and longer.

It seems that when the queue is stuck for some unknown reason on your PC, triggering a write-back, which pauses and un-pauses background processing, gets the queue running again.

Looking at your log file, I see this:

AutoTagger: Aborting because of error [530] 'PTCAIConnectorLMStudio: HTTP error.'

Error 530 means "HTTP error". IMatch uses HTTP to communicate with Ollama and LM Studio. Both offer their services by starting a web server. Apparently,. 530 is not a timeout, but something "else". LM Studio did not respond in a supported way, but also there was no status message with additional information. Else IMatch would have logged it.

AutoTagger cannot community with LM Studio or gets unexpected responses.

It automatically pauses for one minute or so and then tries again. it is not uncommon for cloud-based services like OpenAI to be temporarily unavailable due to overload, and AutoTagger deals with this by waiting a while and then automatically retrying.

I have never seen this with Ollama or LM Studio. Since this seems to affect only you, do you run any specific software? Which virus checker are you using? Did it log anything? Virus checkers might consider LM Studio a treat, because it opens a port and receives "commands"...

I only use Microsoft Defender. I have added the entire Photools folder into excluded, but I have not done the same with either Ollama or LM studio. I will try that. Also my router has very robust security, which I have not checked. I will do that as well

.

[/i]

Do you see the "Files with problems" AutoTagger notification at the bottom of the IMatch window

Not that ive noticed. If I clear the cue I have seen that a few times, but not when it hangs. It just said x number of files added to autotagger category or whatever it says. I do see problems with files that are in the wrong format occasionally, but not when autotagger is running.

[Mario]

Please don't quote so much. Please don't quote in quotes. This makes your post almost impossible to read or to find any useful information you might have added in-between all the nested quotes.

Quote only what's necessary. Thank you.
Like so:

Not that ive noticed. If I clear the cue I have seen that a few times, but not when it hangs. It just said x number of files added to autotagger category or whatever it says. I do see problems with files that are in the wrong format occasionally, but not when autotagger is running.

I need a log file in debug logging mode from an IMatch session where this happened. See log file for details.

[jamiesaun85]

Please don't quote so much. Please don't quote in quotes. This makes your post almost impossible to read or to find any useful information you might have added in-between all the nested quotes.

Quote only what's necessary. Thank you.
Like so:

Not that ive noticed. If I clear the cue I have seen that a few times, but not when it hangs. It just said x number of files added to autotagger category or whatever it says. I do see problems with files that are in the wrong format occasionally, but not when autotagger is running.

I need a log file in debug logging mode from an IMatch session where this happened. See log file for details.

My mistake, Indeed it does flash a yellow square at the bottom and says "files with problems, added to category for review." So I went into that category and the blue info button doesn't tell me what the problem was. Only info about the file itself. This just now happened. It's from a long cue that Autotagger is working on now, and I did not have to write back this time. It's actually working now and doing its thing so far without any issues except the two it skipped and flagged for review. I will get this log to you momentarily.

Let me check again real quick.....

Yep, it's still going. It's up to 150. This morning, at least, I did not have to write back anything. Let me grab that log. It should be near the top, hopefully

[jamiesaun85]

Here it is sir. Thank you for the help

[Mario]

The Orange Warning notification should not flash. It is shown and stays on screen until you close it.
It tells you that AutoTagger ran into a problem (which should be really rare) and that the files causing the problems have been added to the category.

The blue "Info" button in the File Window opens the File Window tip. It does not show error messages.

In the log I see:

W> AutoTagger: Aborting because of error [530]

(HTTP Error, we had that already).

and several times

W> AutoTagger: Aborting because of error 'Failed to load model "gemma-3-12b-it". Error: Error loading model.'

No idea why LM Studio could not load the model on your system. Not enough memory?
Look at the LM Studio logs. Click the "LM Studio" button at the bottom left and use the "Open App Logs" command.

Also one of

UpdateQueue (AutoTagger): Service error 310

which means "Cancelled". Could be caused by many things, like closing the database while AT is running...

[jamiesaun85]

The Orange Warning notification should not flash. It is shown and stays on screen until you close it.
It tells you that AutoTagger ran into a problem (which should be really rare) and that the files causing the problems have been added to the category.

The blue "Info" button in the File Window opens the File Window tip. It does not show error messages.

In the log I see:

W> AutoTagger: Aborting because of error [530]

(HTTP Error, we had that already).

and several times

W> AutoTagger: Aborting because of error 'Failed to load model "gemma-3-12b-it". Error: Error loading model.'

No idea why LM Studio could not load the model on your system. Not enough memory?
Look at the LM Studio logs. Click the "LM Studio" button at the bottom left and use the "Open App Logs" command.

Also one of

UpdateQueue (AutoTagger): Service error 310

which means "Cancelled". Could be caused by many things, like closing the database while AT is running...

Yes sir, if the canccled one was at the end, that was me, I cancled the cue, copied the log and closed IMatch. So if so, disregard that one. Ill focus on the other two, but that will have to be this evening. Gotta go to work.

Thanks again, ill be back I'm sure.

[jamiesaun85]

In briefly scanning the log inside LMStudio, I found this:

[2025-04-09 06:35:02.484] [error] (node:4500) MaxListenersExceededWarning: Possible EventEmitter memory leak detected. 11 close listeners added to [Server]. MaxListeners is 10. Use emitter.setMaxListeners() to increase limit.

Here is the whole log if it helps.

[Mario]

I'm not a LM Studio exert. The log file contains tons of lines with error this and error that. I have many errors in my LM Studio log file, too. But they don't seem to be serious, since LM Studio was working for almost 3 hours last night, processing thousands of AT requests without any issues.

[Stenis]

Before the last update to Version 2025.2.4, Autotagger has had several issues. LLava 7 still works the best, but even it is now doing the same thing. Autotagger just sops working, and how many it will do seems totally random. If I cue up, say 100 files to annalyze and tag, it mwy do anywhere from 5 to 75 or whatever, but eventually it will stick on one and not move from there.

Also, this is with LLava 7b. If I use any other, including LLava 13, it will only do 3 or 4 befoore it stops. Gemma is particullary bad. It will only do 1 to 2 files at  a time and thats it. It will stop working after that and i have to clear the cue and try again.

Another note, when using the best one, LLava 7, it only works better if I dont touch it. I cant scroll and see what its filling out. I cant switch to another page and work on other photos and I cant switch to another app and and do something else on the pc  while its working. If I dont touch it, it will do a couple hundred or so before it chkes up and stops. If i touch it, it will basically stop immediatley and noot work again until I clear cue and run it again.

Before the update, LLava 7 would run an entire cue without ever stopping. I switched to a studio driver but it made no difference.

Is it just me or does anybody else have this problem?

When my AutoTagger might hang I have so far always get it released by initiate a forced update through pressing Shift+Alt+S.

Well that was fast. Dude, that totally worked. You are so awesome! Thank you.

Mario, It hangs, or stops writing anything. I press alt+shift+S and it writes back the pending metadata. I touch nothing else and autotagger starts working and finished the cue. Not sure if this information helps you or not. But Stenis is very much correct. That worked like a charm.

Shift+Alt+S or the tripple push for "Forced Update" has in fact developed to become such a standard procedure in my workflow with Autotagger that I don't even think of when using it. It goes by itself now almost automatic.

I don't accept even the least delay now before I force the update. So It would be good if you can fix this and if nothing else helps you might just add that in the Autotagger procedure. Then we don't have to do it all the time by ourselves.

[Mario]

Start IMatch.
Witch to debug logging (Help > Support).
Run AutoTagger on some files.
If the files are just enqueued, but don't run, make a copy of the log file (Help > Support). Save to disk, ZIP and attach.

Having to force a write-back to make AutoTagger run is ridiculous.

So far, as per the log files, AutoTagger did not run because it received unexpected results or HTTP errors from LM Studio on your system. Or strange issues like "failed to load model", which I have never seen before.
AutoTagger will delay execution indefinitely when such things happen, retrying every minute.

If writing back data changes this somehow, maybe the reason is the virus checker? Maybe it is distracted in that case? I have no idea.

Maybe it's related to the database? Create a new database, index a folder, try AutoTagger. Same result?

[Stenis]

Start IMatch.
Witch to debug logging (Help > Support).
Run AutoTagger on some files.
If the files are just enqueued, but don't run, make a copy of the log file (Help > Support). Save to disk, ZIP and attach.

Having to force a write-back to make AutoTagger run is ridiculous.

So far, as per the log files, AutoTagger did not run because it received unexpected results or HTTP errors from LM Studio on your system. Or strange issues like "failed to load model", which I have never seen before.
AutoTagger will delay execution indefinitely when such things happen, retrying every minute.

If writing back data changes this somehow, maybe the reason is the virus checker? Maybe it is distracted in that case? I have no idea.

Maybe it's related to the database? Create a new database, index a folder, try AutoTagger. Same result?

No it is not ridiculous at all it is a work around to make it work practically users do as long as this bug and all other bugs prevail in both programs I once developed and the ones you are developing now. This is the way people always have tried to cope when developers have overlooked some problems.

Good luck with the debugging. Tell us when you are ready.

[jamiesaun85]

He wasn't telling you that what you are saying is ridiculous. He was saying It's ridiculous to have to do that if it's true.

Which, yes, it is true. At least recently, and in almost all circumstances, but not all. This morning was a good example. I will create a new database with 100 or so files and see what we get there. I'll also copy a log anytime it hangs, before I clear the cue. Sorry about that, not sure why I thought it would help to do that. I am definitely not a software guy

[jamiesaun85]

And it will not surprise me in the least if it turns out to be the database itself. Just last week, I uploaded a log that was half errors, remember? Lol. That was me.

Well see. Nobodies blaming anybody here. Sometimes stuff just happens. Software is not easy, not by a long shot, and we've all got our ares we are proficient in. Mine happens to be DC electrical and electronics diagnostics. NOT software. Don't know the first thing about software.

[jamiesaun85]

Okay, Mario.

I qued 157 files and ran autotagger. It hung after the first file again. So I made a copy of that log. Then I wrote back the files and autotagger started working again. I then made a copy of that log too. Between the two, Maybe there's good information in there. Let's hope so. Thank you for your time.

I added a 1 to the name of the file copied after writing back the metadata, just so you know which is which.

[Mario]

So this is what the logs tell me:

04.09 17:29:07+ 3828 [14A0] 10  I> AutoTagFilesInQueue: [127052] and 157 files.

157 files are added to the AutoTagger processing queue.  A few seconds later, AutoTagger begins processing files:

04.09 17:29:18+    0 [1C6C] 02  I> AutoTagger: AI time: 11093 ms. 92 output tokens, max_tokens: 123.
04.09 17:29:21+    0 [1C6C] 02  I> AutoTagger: AI time: 2813 ms. 99 output tokens, max_tokens: 123.
04.09 17:29:24+    0 [51A8] 02  I> AutoTagger: AI time: 5500 ms. 102 output tokens, max_tokens: 123.

IMatch updates the file history of the 3 files.

04.09 17:29:24+    0 [3720] 10  I> FileHist Delegate: 3 of 3 events written in 0ms

Now there are some cache operations, clearing the preload queue.
Did you have the Quick View Panel or Viewer open?


Then some app requests data from IMWS. GPS Coordinates are loaded. Which apps did you have open at the time?
Map Panel? People View? Event View? 

04.09 17:29:26+    0 [4DE0] 10  M>  <  0 CIMatchWebServices::HandlerGET_IDList
04.09 17:29:26+  15 [4A90] 50  M> >  0 CIMatchWebServices::HandlerGET_Files
04.09 17:29:26+  16 [4A90] 10  M> >  0 PTMetabase::GetGPSCoordinates
04.09 17:29:26+    0 [4A90] 10  M> >  0 CIMSQLite::CreateTemporaryTableOid
04.09 17:29:26+    0 [4A90] 10  M>  <  0 CIMSQLite::CreateTemporaryTableOid
04.09 17:29:26+  297 [4A90] 10  M>  <  0 [297ms] PTMetabase::GetGPSCoordinates
04.09 17:29:26+    0 [4A90] 50  M>  <  0 [313ms] CIMatchWebServices::HandlerGET_Files

Then IMatch wants to refresh the @Keywords category but skips it because the background processing queue is busy.

04.09 17:29:29+    0 [2E68] 10  I> RefreshGroup-Skipped (Queue Busy): '@Keywords'.

Then "nothing happens", except from some background folder scans (caused by Windows sending "modified" messages.

The update queue is enabled, there are no warnings or errors. Looks pretty normal.
The first log ends at:

04.09 17:31:11+30812 [4E20] 02  I> FolderSweeper: processed 52 folders in 0ms. Deepcheck: 0

In the second log, we start at

04.09 17:31:41+29891 [14A0] 10  M> >  0 CIMTaskMetabase::WriteBackAll

when you trigger a write-back. This causes thousands of operations.

7 seconds later, while writeback is still running, AutoTagger starts processing the next file and finishes after ~ 4 seconds.

04.09 17:31:48+    0 [385C] 02  I> AutoTagger: [127175] 'C:\Users\jamie\Pictures\IMatch\Family Pictures\Pets\Carrie 169.jpg'.
04.09 17:31:51+    0 [385C] 02  I> AutoTagger: AI time: 3063 ms. 100 output tokens, max_tokens: 122.

So, in summary:

AutoTagger processes 3 files. Then no more AutoTagger tasks are processed until you start the write back.
Between the last AutoTagger task completed and the pause, the cache is cleared, some app requests file data from IMWS, GPS coordinates are loaded, and @Keywords tries to update but does not since the queue is busy, probably invalidating @Keywords again anyway.

Between 17:29:29 and 04.09 17:31:41 (two minutes+) IMatch is idle. Then you start a write-back.

I don't see anything that tells me why the background processing queue has decided to not schedule AutoTagger tasks for processing. No errors. No warnings. The queue is not paused and enabled.

Which apps were running at 04.09 17:29:26?
Anything else unusual during that time? 

[Mario]

No it is not ridiculous at all it is a work around to make it work practically users do as long as this bug and all other bugs prevail in both programs I once developed and the ones you are developing now. This is the way people always have tried to cope when developers have overlooked some problems.

I wrote that I find it ridiculous that he has to perform a write back just to get AutoTagger running again.
Don't make such assumptions. Keep in mind that English is not my first language.

[jamiesaun85]

Did you have the Quick View Panel or Viewer open?


Then some app requests data from IMWS. GPS Coordinates are loaded. Which apps did you have open at the time?
Map Panel? People View? Event View?


Which apps were running at 04.09 17:29:26?
Anything else unusual during that time?

I'm sorry about this, Mario, I don't yet don't know what most of that is. I've actually only been running IMatch in earnest for about a week or so. As you know better than anybody, it is a complicated piece of software. There is so much to it that it's like drinking from a fire hose. But in a good way, of course. The more it can do, the better, and this is the best digital management software I have ever seen, by far. But It's intense, I'll try to help you out best I can, but as of now I probably only use maybe 5% of its functions.

No to the people view and event view, I think. Yes to the map view, I have that open at all times. I'll post a screenshot of exactly how I have I match set up at all times. Also, no, I didn't notice any other unusual activity at the time. 


Weird, I've never had a forum ask for an insert that way before, not sure what to put. Anyway, that is how I always keep IMatch for now. The only thing that changes is the top right window. I either have it on default metadata, browser metadata or on info and activity. For the main window, I bounce between Media and Folders and People. I haven't started in Events or Categories or anything else. Still just cleaning up my ridiculous database and getting it organized for a change.

[Mario]

The Map Panel being open explains the IMWS requests we see after the 3rd file was processed by AutoTagger.

What I'm trying to find out is why AutpTagger enqueues 157 images, then starts processing them and stalls after the 3rd file. Without any warning, error message or other unusual operation.

When you trigger a write-back, thousands of database operations are performed and "somehow" this causes AutoTagger to continue with the next image. But I have no idea what.

I can see in the log file that the queue is enabled and not paused. All tasks currently enqueued for processing will run in order of priority. Or should. But AutoTagger does not. Or maybe he decided to make a several minute pause. I don't know.

When you, instead of triggering a write back, just run AutoTagger for another image when it is currently not processing, does this have the same effect as the write-back? I mean, causing AutoTagger to start processing again?

Or, reopening the database (just load it again from the Database > Recent) menu?

[jamiesaun85]

When you, instead of triggering a write back, just run AutoTagger for another image when it is currently not processing, does this have the same effect as the write-back? I mean, causing AutoTagger to start processing again?

Or, reopening the database (just load it again from the Database > Recent) menu?

I am not sure, I will try both of those. I was also going to open this same database on one of my other computers and see if the same thing happens or not. I will try all of the above and report back

[thrinn]

Please don't quote so much. Please don't quote in quotes.

@jamiesaun85: If you want to quote just a sentence (or paragraph), you can select the text you want the to quote by dragging the mouse, with left button pressed, over the relevant parts. You will then see an additional button at the bottom "quote selected text".

    

Easier than quoting the whole message and afterwards deleting the parts that are not relevant.
I just want to mention it because it is not obvious as the button is only visible after selecting some text.

[jamiesaun85]

@jamiesaun85: If you want to quote just a sentence (or paragraph), you can select the text you want the to quote by dragging the mouse, with left button pressed, over the relevant parts. You will then see an additional button at the bottom "quote selected text".

Easier than quoting the whole message and afterwards deleting the parts that are not relevant.
I just want to mention it because it is not obvious as the button is only visible after selecting some text.

Hey look at that! it works.

Thank you for the tip, its a good one.

[jamiesaun85]

When you, instead of triggering a write back, just run AutoTagger for another image when it is currently not processing, does this have the same effect as the write-back? I mean, causing AutoTagger to start processing again?

Or, reopening the database (just load it again from the Database > Recent) menu?

So I can answer one of those right now.

Yes it does. If I cue 35 files, and it hangs, and I select another file to tag, yes, that also triggers it to start, and it finished all 35, plus the last one I triggered. So it completes all 36 of them without anymore hanngups or anything. Works like a charm. Same exact effect 

[Mario]

Well, at least that's faster than write-back.

Adding another file to AutoTagger will "signal" the queue (basically saying "Hey, work to do!"). And the AutoTagger thread will be awaken and checking for work.

But I still have no clue why it stops processing while there are still tasks in the queue.

Looking at the code, it logs messages when it aborts for any reason or has to wait and re-schedule when it receives e.g. a timeout from the AI. Or when it runs into a rate limit of sorts.

You said that you get the same effect from both Ollama and LM Studio. So this rules out the AI as the cause for me.

I have not experienced this for myself and don't know where to start searching.
How do you add files to AT? <Alt>+<F7> or <ShiftY>+<Alt>+<F7>?
Does it also stop when you add 1, 2, 3, 4, or 5 files at a time?

I wonder why this seems to affect your installation in such a reproducible manner. Very strange.

[Stenis]

The Map Panel being open explains the IMWS requests we see after the 3rd file was processed by AutoTagger.

What I'm trying to find out is why AutpTagger enqueues 157 images, then starts processing them and stalls after the 3rd file. Without any warning, error message or other unusual operation.

When you trigger a write-back, thousands of database operations are performed and "somehow" this causes AutoTagger to continue with the next image. But I have no idea what.

I can see in the log file that the queue is enabled and not paused. All tasks currently enqueued for processing will run in order of priority. Or should. But AutoTagger does not. Or maybe he decided to make a several minute pause. I don't know.

When you, instead of triggering a write back, just run AutoTagger for another image when it is currently not processing, does this have the same effect as the write-back? I mean, causing AutoTagger to start processing again?

Or, reopening the database (just load it again from the Database > Recent) menu?

I have added maybe 7-8 new Autotagger processes yesterday while there was both pending update transactions and these 7-8 later newer pending AT requests and that did not trigger a new AT processing. I still had to release that lock with the three finger "magic".

Mario you are doing really well with the English as far as i'm concerned, English insn't my native language either :-). The deaper unspoken meanings in the English that are so very obvious and clear for Americans and Brittons are not so for me either but generally tech English is luckily mostly a different thing.

I remember how terrible I found it was to read some course litterature in the university written by Brittish writers. Americans are often far easier for me (had no problems with reading "Macroeconomics" by Lipsey & Steiner). Nobel Prize winner Joseph Brodsky answered a question about the difference writing in US English and Russion and said: When I write in Russian I feel I have make it come out right and beautifully but when writing in English I want it "to make sense".

I hated all these synonyms those Brittish stapled on each others that forced me to sit there all the time with the dictionary. How different ut was despite it was written in the same language of Lipsey& Steiner :-).

[Mario]

What is an AutoTagger process? Are you experiencing the same problem the OP faces? If not, please open a separate bug report. Talking about too many things, unrelated things, in a bug report thread is not helpful.

[jamiesaun85]

Well, at least that's faster than write-back.

I was thinking the exact same thing, LOL. Honestly, I'm no longer worried about it. It was a huge problem, but as we have discovered, there is an easy fix. If it's that easy to get it working, simply adding one more to the cue to make it remember, and it isn't giving any clues as to why it's doing it; then don't even worry about it.

I've just gotten started with Imatch, I've got so much further to go that this is now down the priority list. If it becomes a bigger problem, If it gets worse or something, ill get back to you. I appreciate the time and effort you have put in on this. The personal attention from the man himself is very rare in the software world.

It's a great product, The service is just as good. I'll promote this software to everybody I know.
Keep up the good work.

Jamie

[jamiesaun85]

Mario you are doing really well with the English as far as i'm concerned, English insn't my native language either :-). The deaper unspoken meanings in the English that are so very obvious and clear for Americans and Brittons are not so for me either but generally tech English is luckily mostly a different thing.

Well, don't feel bad. It is the largest and most complicated language that has ever existed, by far. Frankly, I'm always very impressed when anybody gets proficient with English as a second language, and compliment them whenever I notice it. Very impressive.

I'm sure I (could) learn English if I didn't grow up with it and learn it as a native tongue. But I sure wouldn't want to. Its very confusing indeed.

Edit: Oh, and just to add, there is a difference between British style English and American English. Things are spelled differently, the vernacular is also completely different. Just to add a little more confusion, because apparently it wasn't already confusing enough.

[sinus]

OT: Do you speak from the English language?
 Because then I would say, that is surely not the case. But this is only my opinion.