[GH-ISSUE #11485] Model ignores prompt and produces unrelated or nonsensical content—even after /clear (Ollama 0.9.6, Windows) #7585

Closed
opened 2026-04-12 19:40:32 -05:00 by GiteaMirror · 1 comment
Owner

Originally created by @ljrirene on GitHub (Jul 21, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/11485

What is the issue?

🐛 Describe the bug
When I run a model locally (e.g., qwen3:14b) on Windows 10 with Ollama v0.9.6, the model completely ignores a simple prompt like hi. Instead, it jumps into math problems or random content. Even after using /clear, which says it cleared the context, the issue persists. This is clearly model hallucination—not context bleed

🚶‍♂️ To Reproduce
ollama run qwen3:14b
Use /clear to reset conversation context (confirmation message appears).

Send a simple greeting: "hi“
Observe how the model responds with unrelated topics (e.g., starts solving math problems or talking about something else).

Expected behavior
A simple greeting like hi should generate a short response such as Hello! or Hi there!, not unrelated complex content or math.

🌍 Environment

Model: qwen3:14b (GGUF)

I’ve tried /clear, /bye, restarting Ollama, running with /no_think, and even modifying sampling parameters—nothing changed.

OS

Windows 10

GPU

Nvidia A4000

Ollama version

v 0.9.6

Originally created by @ljrirene on GitHub (Jul 21, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/11485 ### What is the issue? 🐛 Describe the bug When I run a model locally (e.g., qwen3:14b) on Windows 10 with Ollama v0.9.6, the model completely ignores a simple prompt like hi. Instead, it jumps into math problems or random content. Even after using /clear, which says it cleared the context, the issue persists. This is clearly model hallucination—not context bleed 🚶‍♂️ To Reproduce ollama run qwen3:14b Use /clear to reset conversation context (confirmation message appears). Send a simple greeting: "hi“ Observe how the model responds with unrelated topics (e.g., starts solving math problems or talking about something else). ✅ Expected behavior A simple greeting like hi should generate a short response such as Hello! or Hi there!, not unrelated complex content or math. 🌍 Environment Model: qwen3:14b (GGUF) **I’ve tried /clear, /bye, restarting Ollama, running with /no_think, and even modifying sampling parameters—nothing changed.** ### OS Windows 10 ### GPU Nvidia A4000 ### Ollama version v 0.9.6
GiteaMirror added the bug label 2026-04-12 19:40:32 -05:00
Author
Owner

@ljrirene commented on GitHub (Jul 21, 2025):

Fixed after adding the following template to modelfile.
Related issue: https://github.com/ollama/ollama/issues/8902
Template link: https://ollama.com/library/qwen3:14b/blobs/ae370d884f10

<!-- gh-comment-id:3098191785 --> @ljrirene commented on GitHub (Jul 21, 2025): Fixed after adding the following template to modelfile. Related issue: https://github.com/ollama/ollama/issues/8902 Template link: https://ollama.com/library/qwen3:14b/blobs/ae370d884f10
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#7585