Number of tokens indicator gone in the new update 0.4.0

[wd357dui]

Which version of LM Studio?
LM Studio 0.4.0 (Build 18)

Which operating system?
Windows 10

What is the bug?
In the previous version, there was an indicator in the lower right corner of the application, which said how many tokens the current chat has, and how full the context window is (in percentage).
But in the version I just upgraded to, 0.4.0, you can see how many tokens one response took by hovering on the clock sign, but the aforementioned indicator in the lower right corner of the application was gone.
This is an important feature in my use case (fiction writing) because I need to budget the entire context window. My input prompt may take thousands of tokens because I need the model to be entirely aware of all previous paragraphs.

Screenshots
The arrow points to the location where the number of tokens indicator should have been.

[yessika-commits]

I just updated to 0.4.0 and first thing I thought was, why are tokens per second information gone. Okay, it's not, just hidden. Fine, I can deal with that.

Then I want to know how much context I've used up after a few messages... and it's just gone. No option to enable or find it anywhere. This is critical when using LMStudio as a chat interface. I know the update is for developers and being able to use parallel messaging, but for my use of LMStudio, this removal has completely crippled it for me and I look for alternatives unfortunately in the mean time. I may just roll back LMStudio to the last version for now.

[Paul-162]

Yes, this is a problem for me on the Windows 11 version. I love the updates (can run QwQ models with quantized KV caches now, literally a 3*speed increase!) but I can't compare the performance directly until I know how many context tokens my chat has.

Also I seem to have lost the ability to specify context limits, such as 'rolling window' or 'stop when max context reached' which is important to me.

Also the 'remember model parameters" doesn't work, every time I load a model I have to reset every parameter in the advanced tab... I figured out how to set the context size (from 'options' now instead of on a per model basis) but even that should be 'per model', not global! Different models use different amounts of VRAM and you should be able to configure them independently!

This is a huge performance increase for me but a massive drop in functionality! Love this software, but this update seems 'half baked', too many changes and not enough backwards compatibility...

Anyway, hope this will be fixed soon, I won't roll back because the performance now rocks, but I need to know how much context I'm using and what happens when that context runs out.

[Paul-162]

Found a work around for now:

Right-click on the conversation, select "Show in File Explorer", open the file (in Notepad on Windows), line 6 in the JSON file is your current 'tokenCount' value (if not, just search for the string 'tokenCount'). You have to reload the JSON each time your conversation changes...

[wd357dui]

I just realized that the "continue generating" button is gone, too! WTH? That was the only effective method I could use to control the writing AI's behavior directly! This is the last straw; I'm going to have to go back to a 0.3.x version...

[Paul-162]

Hey, wd357dui, in case you didn't notice 0.4.1 seems to have fixed all the issues, they've rolled back parts of the interface and context use is back (now a graphical gauge, hover over it to get numerical values), as is the 'continue' button. The only major difference now between 3.x.x is that you can only edit the 'load' parameters (such as K/V quantization and GPU offloading) when a model is unloaded, when a model is loaded you can only change runtime settings, the 'load' tab on the new 'tabbed' model settings doesn't even appear (but runtime changes take effect immediately now). Previously you could edit a model while it was loaded but changes wouldn't take effect until you re-loaded it. 0.4.0 was definitely a miss-step, but 0.4.1 is good again!