[GH-ISSUE #8296] Some Models seem to be crashing while using with JSON Schema mode #82996

Open
opened 2026-05-09 16:54:38 -05:00 by GiteaMirror · 1 comment
Owner

Originally created by @VistritPandey on GitHub (Jan 3, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8296

Originally assigned to: @ParthSareen on GitHub.

What is the issue?

OS: MacOS Sequoia | Linux
Processor | GPU: M3, M4 Pro, i7 with RTX 4070 [Same issue across various devices]
ollama version is 0.5.4

When I try running a batch to get results for it If I specify a format (JSON Schema), then it randomly stops processing randomly at a certain level of the batch. Let's say if I have 100 images, it will randomly stop at 15 or 16 etc. but when you restart the same batch, it does not pause/stop at the number it stopped before. When using Ollama Serve, I am not able to see any errors. But when using without the JSON Schema; it does not fail. This means the issue may be with certain models and how they are handling the JSON Schema.

I used the following models, with vision:
Llama 3.2-vision (never crashed or paused)
Llava-llama3 (crashes/stops occasionally with the JSON Schema)
Minicpm-v (crashes/stops occasionally with the JSON Schema)

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.5.4

Originally created by @VistritPandey on GitHub (Jan 3, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8296 Originally assigned to: @ParthSareen on GitHub. ### What is the issue? OS: MacOS Sequoia | Linux Processor | GPU: M3, M4 Pro, i7 with RTX 4070 [Same issue across various devices] ollama version is 0.5.4 When I try running a batch to get results for it If I specify a format (JSON Schema), then it randomly stops processing randomly at a certain level of the batch. Let's say if I have 100 images, it will randomly stop at 15 or 16 etc. but when you restart the same batch, it does not pause/stop at the number it stopped before. When using Ollama Serve, I am not able to see any errors. But when using without the JSON Schema; it does not fail. This means the issue may be with certain models and how they are handling the JSON Schema. I used the following models, with vision: Llama 3.2-vision (never crashed or paused) Llava-llama3 (crashes/stops occasionally with the JSON Schema) Minicpm-v (crashes/stops occasionally with the JSON Schema) ### OS macOS ### GPU Apple ### CPU Apple ### Ollama version 0.5.4
GiteaMirror added the bug label 2026-05-09 16:54:38 -05:00
Author
Owner

@karkisa commented on GitHub (Nov 14, 2025):

@VistritPandey thanks for opening this issue
@ParthSareen I face this issue as well for different set of models
more so with the qwen2.5 vl
the model just holds on the resource , and re-processing does not really work as it ends up in same situation

<!-- gh-comment-id:3533486123 --> @karkisa commented on GitHub (Nov 14, 2025): @VistritPandey thanks for opening this issue @ParthSareen I face this issue as well for different set of models more so with the qwen2.5 vl the model just holds on the resource , and re-processing does not really work as it ends up in same situation
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#82996