Mintplex-Labs/anything-llm
1
0
Fork 0
mirror of https://github.com/Mintplex-Labs/anything-llm.git synced 2026-07-19 22:23:50 -04:00
Code Issues 328 Packages Projects Releases 10 Wiki Activity

[GH-ISSUE #1841] [BUG]: After about 19 to 20 inputs all of a sudden the context-window get revered? #1198

New Issue
Closed
opened 2026-02-22 18:23:39 -05:00 by yindo · 2 comments
yindo commented 2026-02-22 18:23:39 -05:00
Owner
Copy Link
Copy Source

Originally created by @Dement242 on GitHub (Jul 10, 2024).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/1841

How are you running AnythingLLM?

AnythingLLM desktop app

What happened?

After about 19 to 20 inputs from me it feel like all of a sudden the context-window get revered or something, You can have a normal back and forth, building on the previous messages, but then from one input to the next the model get all confused and start talking about random things from earlier in the conversation, and it is impossible to get it back on track. I have to /reset and start over.

I have checked and my token size is correct for llama-3: 8192, Chat history is set to 20.

Windows AnythingLLM v1.5.8 (I had version 1.5.6 before, and I do not recall it happening in that version)

Are there known steps to reproduce?

Talk to the model, move from subject to subject, after ~20 messages it will not answer what you are talking about now, but some random subject from before (usually from somewhere in the beginning).

Originally created by @Dement242 on GitHub (Jul 10, 2024). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/1841 ### How are you running AnythingLLM? AnythingLLM desktop app ### What happened? After about 19 to 20 inputs from me it feel like all of a sudden the context-window get revered or something, You can have a normal back and forth, building on the previous messages, but then from one input to the next the model get all confused and start talking about random things from earlier in the conversation, and it is impossible to get it back on track. I have to /reset and start over. I have checked and my token size is correct for llama-3: 8192, Chat history is set to 20. Windows AnythingLLM v1.5.8 (I had version 1.5.6 before, and I do not recall it happening in that version) ### Are there known steps to reproduce? Talk to the model, move from subject to subject, after ~20 messages it will not answer what you are talking about now, but some random subject from before (usually from somewhere in the beginning).
yindo added the possible bug label 2026-02-22 18:23:39 -05:00
yindo closed this issue 2026-02-22 18:23:39 -05:00
yindo commented 2026-02-22 18:23:40 -05:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (Jul 10, 2024):

All models have limited context windows, so at a certain point, we need to truncate or abbreviate messages from the chat window. This is just the LLM's chat history being managed in the background and especially going from subject to subject information will be lost during truncation

@timothycarambat commented on GitHub (Jul 10, 2024): All models have limited context windows, so at a certain point, we need to truncate or abbreviate messages from the chat window. This is just the LLM's chat history being managed in the background and especially going from subject to subject information will be lost during truncation
yindo commented 2026-02-22 18:23:40 -05:00
Author
Owner
Copy Link
Copy Source

@Dement242 commented on GitHub (Jul 10, 2024):

Yes, that I understand. And I made a small spelling error in the title, it should be reversed, not revered. What I expect to happen is that oldest messages get pushed out, but (as I can not see the full context that get sent to the model) I suspect every time it start to truncate at 20 message history, the model tries to answer my input and treat the absolute first (or second) message as if it was the latest message.

@Dement242 commented on GitHub (Jul 10, 2024): Yes, that I understand. And I made a small spelling error in the title, it should be reversed, not revered. What I expect to happen is that oldest messages get pushed out, but (as I can not see the full context that get sent to the model) I suspect every time it start to truncate at 20 message history, the model tries to answer my input and treat the absolute first (or second) message as if it was the latest message.
yindo changed title from [BUG]: After about 19 to 20 inputs all of a sudden the context-window get revered? to [GH-ISSUE #1841] [BUG]: After about 19 to 20 inputs all of a sudden the context-window get revered? 2026-06-05 14:39:29 -04:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
Desktop
Docker
Integration Request
Integration Request
OS: Linux
OS: Mobile
OS: Windows
UI/UX
blocked
bug
bug
core-team-only
documentation
duplicate
embed-widget
enhancement
feature request
github_actions
good first issue
investigating
needs info / can't replicate
possible bug
pull-request
Mirrored from GitHub Pull Request
 question
stage: specifications
wontfix
No Label possible bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: Mintplex-Labs/anything-llm#1198
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?