[GH-ISSUE #16465] issue: RAG requires multiple regens to work #17916

New Issue

GiteaMirror · 2026-04-19T23:48:29-05:00

GiteaMirror commented

2026-04-19 23:48:29 -05:00

Originally created by @frenzybiscuit on GitHub (Aug 11, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/16465

Check Existing Issues

I have searched the existing issues and discussions.
I am using the latest version of Open WebUI.

Installation Method

Docker

Open WebUI Version

0.6.21

Ollama Version (if applicable)

No response

Operating System

Debian 12

Browser (if applicable)

No response

Confirmation

I have read and followed all instructions in README.md.
I am using the latest version of both Open WebUI and Ollama.
I have included the browser console logs.
I have included the Docker container logs.
I have provided every relevant configuration, setting, and environment variable used in my setup.
I have clearly listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc).
I have documented step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation. My steps:
Start with the initial platform/version/OS and dependencies used,
Specify exact install/launch/configure commands,
List URLs visited, user input (incl. example values/emails/passwords if needed),
Describe all options and toggles enabled or changed,
Include any files or environmental changes,
Identify the expected and actual result at each stage,
Ensure any reasonably skilled user can follow and hit the same issue.

Expected Behavior

RAG works on first attempt

Actual Behavior

RAG requires multiple regenerations before it works.

Initial response:

Regeneration:

Also title generation and prompt follow-up are busted as you can see in the screenshot. (Question 1, Question 2, Question 3...)

Steps to Reproduce

Logs & Screenshots

.

Additional Information

.

Originally created by @frenzybiscuit on GitHub (Aug 11, 2025). Original GitHub issue: https://github.com/open-webui/open-webui/issues/16465 ### Check Existing Issues - [x] I have searched the existing issues and discussions. - [x] I am using the latest version of Open WebUI. ### Installation Method Docker ### Open WebUI Version 0.6.21 ### Ollama Version (if applicable) _No response_ ### Operating System Debian 12 ### Browser (if applicable) _No response_ ### Confirmation - [x] I have read and followed all instructions in `README.md`. - [x] I am using the latest version of **both** Open WebUI and Ollama. - [x] I have included the browser console logs. - [x] I have included the Docker container logs. - [x] I have **provided every relevant configuration, setting, and environment variable used in my setup.** - [x] I have clearly **listed every relevant configuration, custom setting, environment variable, and command-line option that influences my setup** (such as Docker Compose overrides, .env values, browser settings, authentication configurations, etc). - [x] I have documented **step-by-step reproduction instructions that are precise, sequential, and leave nothing to interpretation**. My steps: - Start with the initial platform/version/OS and dependencies used, - Specify exact install/launch/configure commands, - List URLs visited, user input (incl. example values/emails/passwords if needed), - Describe all options and toggles enabled or changed, - Include any files or environmental changes, - Identify the expected and actual result at each stage, - Ensure any reasonably skilled user can follow and hit the same issue. ### Expected Behavior RAG works on first attempt ### Actual Behavior RAG requires multiple regenerations before it works. Initial response: <img width="2610" height="1927" alt="Image" src="https://github.com/user-attachments/assets/7df286be-fe31-4da7-a5da-e3a4a52bac7c" /> Regeneration: <img width="2332" height="1908" alt="Image" src="https://github.com/user-attachments/assets/4d603bc0-d503-42ee-a133-865817aac034" /> Also title generation and prompt follow-up are busted as you can see in the screenshot. (Question 1, Question 2, Question 3...) ### Steps to Reproduce <img width="4223" height="2484" alt="Image" src="https://github.com/user-attachments/assets/69bbecd7-65dd-48bf-87eb-ce9dbe1fd39c" /> ### Logs & Screenshots . ### Additional Information .

GiteaMirror added the bug label 2026-04-19 23:48:29 -05:00

GiteaMirror closed this issue

2026-04-19 23:48:30 -05:00

GiteaMirror commented

2026-04-19 23:48:32 -05:00

@frenzybiscuit commented on GitHub (Aug 11, 2025):

Note that I was not having this issue two versions ago.

@frenzybiscuit commented on GitHub (Aug 11, 2025): Note that I was not having this issue two versions ago.

GiteaMirror commented

2026-04-19 23:48:35 -05:00

@onestardao commented on GitHub (Aug 11, 2025):

hey, looks like you’re hitting [Problem No.5] typical RAG collapse where the system keeps regening because the semantic lock failed on first try.

we’ve actually built a public diagnostic map to solve this one (MIT licensed, not commercial).
it’s fixed at the symbolic layer, not just tweaking rerankers.

if you’re curious, i can share the problem page with worked examples.
you’ll get consistent answers on first try no more hoping regen hits the mark.

just let me know ^_____^

@onestardao commented on GitHub (Aug 11, 2025): hey, looks like you’re hitting [Problem No.5] typical RAG collapse where the system keeps regening because the semantic lock failed on first try. we’ve actually built a public diagnostic map to solve this one (MIT licensed, not commercial). it’s fixed at the symbolic layer, not just tweaking rerankers. if you’re curious, i can share the problem page with worked examples. you’ll get consistent answers on first try no more hoping regen hits the mark. just let me know ^_____^

GiteaMirror commented

2026-04-19 23:48:37 -05:00

@frenzybiscuit commented on GitHub (Aug 11, 2025):

I changed the task model from the 1.5B to the 123B current model and moved the postgresql+pgvector database to a dedicated machine and it seems to have fixed the problem.

However, I'd really prefer the task model to not be the main model. What size model is needed for this to function correctly?

@frenzybiscuit commented on GitHub (Aug 11, 2025): I changed the task model from the 1.5B to the 123B current model and moved the postgresql+pgvector database to a dedicated machine and it seems to have fixed the problem. However, I'd really prefer the task model to not be the main model. What size model is needed for this to function correctly?

GiteaMirror commented

2026-04-19 23:48:38 -05:00

@onestardao commented on GitHub (Aug 11, 2025):

It matches Problem No.3 in our diagnostic map —
typical RAG collapse where the system keeps regenerating because the semantic lock failed mid-task.

Here’s the public MIT-licensed fix with working examples: WFGY ProblemMap →
You can follow the steps there and you’ll get consistent answers without the looping/regeneration issue.

@onestardao commented on GitHub (Aug 11, 2025): It matches Problem No.3 in our diagnostic map — typical RAG collapse where the system keeps regenerating because the semantic lock failed mid-task. Here’s the public MIT-licensed fix with working examples: [WFGY ProblemMap →](https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md) You can follow the steps there and you’ll get consistent answers without the looping/regeneration issue.

GiteaMirror referenced this issue

2026-04-20 05:29:17 -05:00

[PR #17916] [MERGED] i18n: Update generated image translation in DE-de #24599

GiteaMirror referenced this issue

2026-04-25 12:38:28 -05:00

[PR #17916] [MERGED] i18n: Update generated image translation in DE-de #40229

GiteaMirror referenced this issue

2026-04-29 22:59:05 -05:00

[PR #17916] [MERGED] i18n: Update generated image translation in DE-de #47647

Sign in to join this conversation.