lmstudio 2025-09-06 17:59:27 [ERROR] The model has crashed without additional information. (Exit code: 18446744072635812000). Error Data: n/a, Additional Data: n/a

[akierum]

Describe the bug
when using lmstudio latest version 0.3.25 with OpenAI like endpoint

2025-09-06 17:59:27 [ERROR]
The model has crashed without additional information. (Exit code: 18446744072635812000). Error Data: n/a, Additional Data: n/a

To Reproduce
Steps to reproduce the behavior:

User model Qwen3-30B-A3B-Instruct-Coder-480B-Distill-v2-Q8_0.gguf
Setup context 190k
try to use the goose, watch it crash very fast.

Expected behavior
working normally

Please provide following information:

• OS & Arch: win11

[cbruyndoncx]

I am confused, you need to provide more detailed steps on how you setup lm studio in goose (which provider you used, streaming or not etc) and what you did exactly or provide a video for us to watch what happens.
What is the developer log in lm studio ?

[akierum]

This is my setup for LMstudio.

Win11
512GB RAM
2x 3090 48GB VRAM

[cbruyndoncx]

Thé URL needs to be the v1/chat/completions endpoint i think.
I have an outstanding pr with changes. Also only 1 custom provider could work as the env var is not provider specific.

Manually editing config.yaml and custom providers you might be Anke to fix, otherwise wait for bugfix in UI

[akierum]

Thé URL needs to be the v1/chat/completions endpoint i think. I have an outstanding pr with changes. Also only 1 custom provider could work as the env var is not provider specific.

Manually editing config.yaml and custom providers you might be Anke to fix, otherwise wait for bugfix in UI

I don't know works without the v1/chat/completions.

[lustfeind]

I get this with nearly every qwen3 coder 30b MLX. Unified RAM is 96 GB, Context size 100000.

[akierum]

The problem is still yet to be identified, reducing context in my case to 188k reduced the frequency of this crash, but it may get frequent for no reason, after restoring PC from sleep this is more frequent. So shift restart for a fresh start on win 11. I wonder if this happens on linux

[michaelneale]

@akierum we wont' really be able to reproduce this, but when you say crash - do you mean the goose cli, the desktop app - the LLM? (that error looks like it is from LMstudio, not goose).

[akierum]

I mean the model crashes, and is some sense even restarting LMstudio gives same error after a while.
Enable Flash attention and KV cache.

[DOsinga]

if the model crashes and it also gives LMStudio trouble, isn't the problem on that side?

[akierum]

The problem is model crashes in LMstudio, but Lmstudio never fixes this problem, I've opened bugfix on their github.