Mintplex-Labs/anything-llm
1
0
Fork 0
mirror of https://github.com/Mintplex-Labs/anything-llm.git synced 2026-07-19 14:13:35 -04:00
Code Issues 328 Packages Projects Releases 10 Wiki Activity

[GH-ISSUE #5580] [BUG]: AnythingLLM as Native Provider ⌛ Timeout Auto-Unload Model too quick #5178

New Issue
Closed
opened 2026-06-05 14:52:29 -04:00 by yindo · 1 comment
yindo commented 2026-06-05 14:52:29 -04:00
Owner
Copy Link
Copy Source

Originally created by @AlonDan on GitHub (May 5, 2026).
Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5580

How are you running AnythingLLM?

AnythingLLM desktop app

What happened?

I've tested same models in different providers, I found out these things:

✅ AnythingLLM as native provider = The fastest (tokens per second), most stable.
❌ LM Studio as provider = Slower (tokens per second), sometimes forgets to answer needs a wakeup "Hello? are you still there?" to remind it to continue or post it's task / answer. it is very random.

THE ISSUE:

In short:
❌ auto-unload is too fast, there is no way to control or change it currently in v1.12.1
❌ continue thread/chat will always have some "pre-delay" time before the actual model re-load again it takes extra minute or so.
❌ I had the same issue with different models: the quick auto-unload with no control is the problem, not the models.

It seems like roughly about 1-2 minutes, using the Native Provider it auto-unload whatever Model I used.
This is extremely inefficient when I want to do other things in this minute.
Then in the next time I press ENTER on the chat for the next question/task, it's not immediately LOADING the model, there is some delay before it actually loading the model, so overall about 1-2 minute (with my specs) until the model is loaded again, this is impossible to work in a nice smooth session flow without these waiting times.

Currently on v1.12.1 there are no options to change the time of the model to stay loaded or at least change to custom X minutes / hours until it auto-unload.

I noticed that if I choose OLLAMA as a provider I do have auto-unload options, even some nice presets.
So I did suggest this: https://github.com/Mintplex-Labs/anything-llm/issues/5571

SOLUTION:

✅ High Priority: Add auto-unload model option similar to how you did for Ollama Provider option: Preset of X Hours / Minutes / Never / custom will be nice. more details and visual examples here: https://github.com/Mintplex-Labs/anything-llm/issues/5571

✅ Low Priority: Get rid of that delay when re-load the model to continue working with it after it unloaded.

LOCAL SPECS TESTED:

• Intel Core Ultra 9 285K
• Nvidia RTX 5090 32 GB VRAM
• 96 GB RAM 6400 Mhz
• Nvme SSD

• Windows 11 Pro
• AnythingLLM - v1.12.1
• LM Studio - v0.4.12

Are there known steps to reproduce?

1 - Use the native AnythingLLM as a provider
2 - Select one of the models (I tried different models, see the list below)
3 - Start a thread, give it a task or talk to it ask something - This will load the model for the first time.
4 - Wait for 1-2 minutes... or check your emails, make a coffee

RESULTS:

  • It will auto-unload the model
  • Continue the chat, you will have to wait about 1-2 minute not just because of loading which is pretty fast, because there is some delay before loading, not sure why, compare to other providers: Ollama / LM Studio there immediately loading the model without pre-delay.

LLM Provider & Model (if applicable)

  • AnythingLLM / Gemma-4-31B Instruct Q4_K_M
  • AnythingLLM / Qwen 3.5 9B
  • AnythingLLM / Qwen 3.5 27B
  • AnythingLLM / Qwen 3.5 35B
  • LM Studio / Gemma-4-31B Instruct Q4_K_M
  • LM Studio / Qwen 3.6 9B Q4_K_M
  • LM Studio / Qwen 3.5 27B Q4_K_M
  • LM Studio / Qwen 3.5 35B-A3B Q4_K_M
  • LM Studio / Qwen 3.6 27B Q4_K_M
  • LM Studio / Qwen 3.6 35B-A3B Q4_K_M

NOTE:

I wanted to test also: Qwen 3.6 but it doesn't appear in the native list in the current version 112.1

Embedder Provider & Model (if applicable)

No response

Originally created by @AlonDan on GitHub (May 5, 2026). Original GitHub issue: https://github.com/Mintplex-Labs/anything-llm/issues/5580 ### How are you running AnythingLLM? AnythingLLM desktop app ### What happened? I've tested same models in different providers, I found out these things: ✅ AnythingLLM as native provider = The fastest (tokens per second), most stable. ❌ LM Studio as provider = Slower (tokens per second), sometimes forgets to answer needs a wakeup "Hello? are you still there?" to remind it to continue or post it's task / answer. it is very random. --- ### THE ISSUE: In short: ❌ auto-unload is too fast, there is no way to control or change it currently in v1.12.1 ❌ continue thread/chat will always have some "pre-delay" time before the actual model re-load again it takes extra minute or so. ❌ I had the same issue with different models: the quick auto-unload with no control is the problem, not the models. It seems like roughly about 1-2 minutes, using the **Native Provider** it auto-unload whatever Model I used. This is extremely inefficient when I want to do other things in this minute. Then in the next time I press ENTER on the chat for the next question/task, it's not immediately LOADING the model, there is some delay before it actually loading the model, so overall about 1-2 minute (with my specs) until the model is loaded again, this is impossible to work in a nice smooth session flow without these waiting times. Currently on **v1.12.1** there are no options to change the time of the model to stay loaded or at least change to custom X minutes / hours until it auto-unload. I noticed that if I choose **OLLAMA** as a provider I do have auto-unload options, even some nice presets. So I did suggest this: https://github.com/Mintplex-Labs/anything-llm/issues/5571 --- ### SOLUTION: ✅ **High Priority**: Add auto-unload model option similar to how you did for Ollama Provider option: Preset of X Hours / Minutes / Never / custom will be nice. more details and visual examples here: https://github.com/Mintplex-Labs/anything-llm/issues/5571 ✅ **Low Priority**: Get rid of that delay when re-load the model to continue working with it after it unloaded. --- ### LOCAL SPECS TESTED: • Intel Core Ultra 9 285K • Nvidia RTX 5090 32 GB VRAM • 96 GB RAM 6400 Mhz • Nvme SSD • Windows 11 Pro • AnythingLLM - v1.12.1 • LM Studio - v0.4.12 ### Are there known steps to reproduce? 1 - Use the native AnythingLLM as a provider 2 - Select one of the models (I tried different models, see the list below) 3 - Start a thread, give it a task or talk to it ask something - This will load the model for the first time. 4 - Wait for **1-2** minutes... or check your emails, make a coffee --- ### RESULTS: - It will auto-unload the model - Continue the chat, you will have to wait about 1-2 minute not just because of loading which is pretty fast, because there is some delay before loading, not sure why, compare to other providers: Ollama / LM Studio there immediately loading the model without pre-delay. ### LLM Provider & Model (if applicable) - AnythingLLM / Gemma-4-31B Instruct Q4_K_M - AnythingLLM / Qwen 3.5 9B - AnythingLLM / Qwen 3.5 27B - AnythingLLM / Qwen 3.5 35B --- - LM Studio / Gemma-4-31B Instruct Q4_K_M - LM Studio / Qwen 3.6 9B Q4_K_M - LM Studio / Qwen 3.5 27B Q4_K_M - LM Studio / Qwen 3.5 35B-A3B Q4_K_M - LM Studio / Qwen 3.6 27B Q4_K_M - LM Studio / Qwen 3.6 35B-A3B Q4_K_M --- ### NOTE: I wanted to test also: **Qwen 3.6** but it doesn't appear in the native list in the current version 112.1 ### Embedder Provider & Model (if applicable) _No response_
yindo added the possible bug label 2026-06-05 14:52:29 -04:00
yindo closed this issue 2026-06-05 14:52:29 -04:00
yindo commented 2026-06-05 14:52:29 -04:00
Author
Owner
Copy Link
Copy Source

@timothycarambat commented on GitHub (May 11, 2026):

Moving to #5571

<!-- gh-comment-id:4426280971 --> @timothycarambat commented on GitHub (May 11, 2026): Moving to #5571
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#5178
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?