frob

Found it, https://ollama.com/library/qwen2.5:72b-instruct-q4_K_S

If it's affecting multiple models of different architectures and different sizes, it may be a hardware issue. #8235 had a similar problem with repeating '@'s, running a [GPU VRAM tester](...

It's a limited tester, perhaps [`OCCT`](https://www.ocbase.com/occt/personal`) might reveal more.

> Using ollama through CLI does not seem to trigger the issue. Whihch way do use the CLI: ```console C:\> olllama run qwen2.5:72b-instruct-q4_K_S Hello Hello @@@ ``` or ```console C:\>...

The latter uses `/api/chat`, the log shows the client used `/api/generate`. What was the client at that time? Open-webui uses `/api/chat` so if you've seen this in open-webui then the...

Some other are reporting the same problem. #9190, #9225 No resolution as yet.

Great! What version of the driver did you upgrade to?

``` #!/bin/bash # Move models around. Example: create an archive of qwen2.5 # $ oarc qwen2.5:0.5b qwen2.5.model # Copy the model to a new machine # $ scp qwen2.5.model remote:/tmp...

```sh #!/bin/bash OLLAMA_MODELS=${OLLAMA_MODELS:-/usr/share/ollama/.ollama/models} cd $OLLAMA_MODELS/manifests/registry.ollama.ai/library || exit for f in $* ; do s=${f##*/} s=${s#sha256?} find . -type f -size -10000c | xargs grep -l "$s" | sed -e 's@^\./@@'...

Port number change is irrelevant. One of the clients is setting `"use_mlock":true` in the API call. A change in model parameters causes a model reload. Either configure the one client...