slow Prompt processing progress

[akierum]

Which version of LM Studio?
Example: LM Studio 0.3.32

Which operating system?
Win11

What is the bug?

Just to process "hello" in Roocode it takes this slow. Full GPU offload.
2x 3090 GPU, Win 11, 512GB RAM
2025-11-14 03:14:05 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 0.0%
2025-11-14 03:14:08 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 5.8%
2025-11-14 03:14:10 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 11.6%
2025-11-14 03:14:13 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 17.5%
2025-11-14 03:14:16 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 23.3%
2025-11-14 03:14:19 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 29.1%
2025-11-14 03:14:22 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 34.9%
2025-11-14 03:14:25 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 40.7%
2025-11-14 03:14:27 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 46.6%
2025-11-14 03:14:30 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 52.4%
2025-11-14 03:14:33 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 58.2%
2025-11-14 03:14:37 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 64.0%
2025-11-14 03:14:41 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 69.8%
2025-11-14 03:14:44 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 75.7%
2025-11-14 03:14:47 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 81.5%
2025-11-14 03:14:50 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 87.3%
2025-11-14 03:14:53 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 93.1%
2025-11-14 03:14:56 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 99.0%
2025-11-14 03:14:56 [INFO]
[Mungert/KAT-Dev-GGUF/Mungert KAT-Dev-q8_0.gguf] Prompt processing progress: 100.0%

Screenshots

Woks fine with Jan.ai, worked fine before with LMstudio too.

[kkjinping]

Me too

[akierum]

Me too

Have you any ideas how to fix it? What GPU you have? I use RTX 3090

[MaxKotlan]

Also have this issue using devstral-small-2507-mlx. Huge bottleneck for me

[zimdin12]

Still an issue.. 4090 and qwen3-coder reap 25b

[akierum]

Tried latest LMstudio v0.3.35 still shows these lines, did not show them before ! WTF? I'm using Jan.ai as alternative more and more now.

[softyoda]

Exact same issue, do i need to setup jan ai to solve the issue?

[anand-kamble]

Facing the same issue on Mac

[Lionad-Morotar]

在Mac上也遇到同样的问题

[artifactdev]

Actually I'm seeing this as well on M4 Pro im combination with roocode in VSCode

[chrislawso]

same issue on machine with 3090ti. other machines with rtx 3060 or other gpu do not exhibit this issue. the issue usually doesnt manifest immediately after LM studio app startup, i can usually run a few prompts at faster speeds and then later the prompt processing slows to a crawl pace.

[akierum]

I wonder how it is not FIXED YET. It's been a YEAR

[samheuck]

Does anyone have any ideas as to the cause?

I'm experiencing this as well. However, the slowdown occurs after a certain amount of inactivity. It feels like MacOS is caching something temporarily, evicting it, then when a new prompt comes it, it has to re-cache everything. I have no evidence of this other than it seems to occur given enough time between prompts.

[Nagal-zn]

Same question with M4pro 64G Mac mini, vscode + roo code, qwen3-coder-30b-8bit

[chaoming]

When I use the chat window, it is working fine. However, when I use Clawdbot/Moltbot with LMStudio, it occurs. I can see a massive JSON prompt from Clawdbot even I ask a question with a few words. I wonder if the massive JSON prompt is causing it.