Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

chat: fix issues with quickly switching between multiple chats #2343

Merged
merged 12 commits into from
May 15, 2024

Conversation

cebtenzzre
Copy link
Member

@cebtenzzre cebtenzzre commented May 13, 2024

If you switch between chats too quickly, it can result in the UI getting into inconsistent states (e.g. a progress bar stuck at 60%) and sometimes seemingly deadlocking (which happens during the new "Waiting for model" status).

To fix UI inconsistencies:

  • Move trySwitchContextInProgress, modelLoadingPercentage, and isCurrentlyLoading properties from the window to the chat so the load status always reflects the current chat
  • Initialize load status at the beginning of loadModel instead of only in setModelInfo - reloadModel calls this too
  • If loadModel is cancelled, quickly return instead of trying to fall back and reporting an "invalid model" error
  • On load failure, instead of showing only the bottom "reload" button, show neither

To prevent quickly switching chats from queueing many slow operations:

  • Skip context switch if an unload is pending
  • Skip unnecessary calls to LLModel::saveState

UI enhancement:

  • Show "waiting for model" as a separate state when switching context, so it's clearer what is taking so long

Other bugfix:

  • Use std::unique_ptr in the LLModelStore and fix a related memory leak - not absolutely necessary for this PR, but it makes some other changes slightly simpler

Non-functional changes:

  • Use std::optional to store zero or one items in the LLModelStore
  • Remove m_shouldTrySwitchContext, which didn't do anything useful, and simplify the related code

@cebtenzzre cebtenzzre force-pushed the fix-chat-switching branch 2 times, most recently from f3cbd09 to bcfcb9c Compare May 14, 2024 17:59
@cebtenzzre cebtenzzre marked this pull request as ready for review May 14, 2024 17:59
@cebtenzzre cebtenzzre requested a review from manyoso May 14, 2024 17:59
@cebtenzzre cebtenzzre marked this pull request as draft May 14, 2024 18:26
@cebtenzzre cebtenzzre marked this pull request as ready for review May 14, 2024 19:11
There is no reason for reloadModel to show slighly different status
information from setModelInfo.

Signed-off-by: Jared Van Bortel <[email protected]>
This fixes a memory leak if there was a model in the store on exit, e.g.
if the user loads a model and then switches to a chat associated with a
different model file without loading it.

ChatLLM::destroyStore is added to fix a heap-use-after-free caused by
the unique_ptr being freed too late. Global destructors are hard.

Signed-off-by: Jared Van Bortel <[email protected]>
When switching context, this state can take a significant amount of
time. Separate it out so the user is less likely to think GPT4All is
completely stuck.

Signed-off-by: Jared Van Bortel <[email protected]>
This aligns its behavior with the upper "reload" button.

Signed-off-by: Jared Van Bortel <[email protected]>
The boolean flag doesn't do anything useful.

Signed-off-by: Jared Van Bortel <[email protected]>
@cebtenzzre cebtenzzre changed the title chat: trying to fix issues with quickly switching chats chat: fix issues with quickly switching between multiple chats May 14, 2024
@manyoso
Copy link
Collaborator

manyoso commented May 15, 2024

This is tricky code. Let's do a huddle for review on this one.

@cebtenzzre cebtenzzre merged commit 7e1e00f into main May 15, 2024
6 of 10 checks passed
@cebtenzzre cebtenzzre deleted the fix-chat-switching branch May 15, 2024 18:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants