[GH-ISSUE #9778] Stuck "Thinking..." Indicator after Canceling Generation #15644

New Issue

GiteaMirror · 2026-04-19T21:48:33-05:00

GiteaMirror commented

2026-04-19 21:48:33 -05:00

Originally created by @liu-hanyu-97 on GitHub (Feb 11, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/9778

Bug Report: Stuck "Thinking..." Indicator after Canceling Generation - Version 0.5.10, llama.cpp

Title: Stuck "Thinking..." Indicator after Canceling Generation

Description:

When interacting with the model, and it enters the "Thinking..." state (indicated by the "Thinking..." logo/indicator), if the generation process is canceled during this "Thinking..." phase, the "Thinking..." logo/indicator remains visible indefinitely. This indicator persists even after the generation is canceled and the model is no longer actively processing. This is a UI issue and does not block further usage, but it is misleading.

Steps to reproduce:

Start a conversation with the model by asking a question or providing a prompt.
Observe that the "Thinking..." indicator (a circular loading icon with "Thinking...") appears next to the model's name while it generates a response.
Click the "cancel" button while the model is actively in the "Thinking..." state.
Observe that the "Thinking..." indicator remains visible even after the generation is cancelled and the model has stopped responding. It does not disappear or revert to a normal state.

Expected behavior:

After canceling the generation process, the "Thinking..." indicator should disappear promptly, indicating that the model is no longer processing and is ready for the next interaction. The UI should clearly reflect that the model is idle and awaiting new input.

Actual behavior:

The "Thinking..." indicator remains stuck and visible indefinitely after canceling the generation. It does not clear even after a significant delay.

Impact:

While this bug is primarily a UI issue and does not prevent core functionality (starting new conversations, regenerating responses), it creates a confusing user experience. The persistent "Thinking..." indicator incorrectly suggests that the model is still processing or busy, when in fact it is idle and awaiting new input. This can lead to user confusion and a perception of unresponsiveness.

Environment:

Version: 0.5.10
Inference Framework: llama.cpp
Browser: [Chrome]
Operating System: [macOS]

Screenshot:

Additional Notes:

This issue does not prevent further usage of the application. Users can still initiate new dialogues or regenerate responses despite the stuck indicator.
The stuck indicator only seems to occur when canceling generation during the "Thinking..." phase. If the model completes generation and then a new prompt is given, the "Thinking..." indicator behaves as expected.

Originally created by @liu-hanyu-97 on GitHub (Feb 11, 2025). Original GitHub issue: https://github.com/open-webui/open-webui/issues/9778 ## Bug Report: Stuck "Thinking..." Indicator after Canceling Generation - Version 0.5.10, llama.cpp **Title:** Stuck "Thinking..." Indicator after Canceling Generation **Description:** When interacting with the model, and it enters the "Thinking..." state (indicated by the "Thinking..." logo/indicator), if the generation process is canceled during this "Thinking..." phase, the "Thinking..." logo/indicator remains visible indefinitely. This indicator persists even after the generation is canceled and the model is no longer actively processing. This is a UI issue and does not block further usage, but it is misleading. **Steps to reproduce:** 1. Start a conversation with the model by asking a question or providing a prompt. 2. Observe that the "Thinking..." indicator (a circular loading icon with "Thinking...") appears next to the model's name while it generates a response. 3. Click the "cancel" button while the model is actively in the "Thinking..." state. 4. Observe that the "Thinking..." indicator remains visible even after the generation is cancelled and the model has stopped responding. It does not disappear or revert to a normal state. **Expected behavior:** After canceling the generation process, the "Thinking..." indicator should disappear promptly, indicating that the model is no longer processing and is ready for the next interaction. The UI should clearly reflect that the model is idle and awaiting new input. **Actual behavior:** The "Thinking..." indicator remains stuck and visible indefinitely after canceling the generation. It does not clear even after a significant delay. **Impact:** While this bug is primarily a UI issue and does not prevent core functionality (starting new conversations, regenerating responses), it creates a confusing user experience. The persistent "Thinking..." indicator incorrectly suggests that the model is still processing or busy, when in fact it is idle and awaiting new input. This can lead to user confusion and a perception of unresponsiveness. **Environment:** * **Version:** 0.5.10 * **Inference Framework:** llama.cpp * **Browser:** [Chrome] * **Operating System:** [macOS] **Screenshot:** ![Image](https://github.com/user-attachments/assets/edd38195-5e73-44b7-b619-e298add17007) **Additional Notes:** * This issue does not prevent further usage of the application. Users can still initiate new dialogues or regenerate responses despite the stuck indicator. * The stuck indicator only seems to occur when canceling generation *during* the "Thinking..." phase. If the model completes generation and then a new prompt is given, the "Thinking..." indicator behaves as expected.

GiteaMirror closed this issue

2026-04-19 21:48:33 -05:00

GiteaMirror commented

2026-04-19 21:48:35 -05:00

@tjbck commented on GitHub (Feb 11, 2025):

Intended behaviour here.

@tjbck commented on GitHub (Feb 11, 2025): Intended behaviour here.

GiteaMirror referenced this issue

2026-04-20 06:30:49 -05:00

[PR #22090] [CLOSED] feat: admin endpoint for direct connections provisioning #26474

GiteaMirror referenced this issue

2026-04-25 14:06:30 -05:00