github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 16:40:08 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #15147] Problem with submitting multiple images to qwen3.5:27b as opposed to qwen3.5:397b-cloud #71757

New Issue
Open
opened 2026-05-05 02:27:22 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-05-05 02:27:22 -05:00
Owner
Copy Link

Originally created by @fchahun on GitHub (Mar 30, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/15147

What is the issue?

I am facing a problem with qwen3.5:27b when submitting a list of 2-3 images (representing the pages of a same document) along with a single text prompt requiring content extraction of the images. With qwen3.5:397b-cloud, everything runs fine as expected. But with qwen3.5:27b, using exactly the same prompt and same images, I consistently get the following error message: "Internal Server Error -- Failed to create new sequence: SameBatch may not be specified within numKeep (index: 3 numKeep: 4 SameBatch: 2017)". This is true whether using the "generate" or "chat" API endpoints.
I am using Ollama 0.18.0 running on an NVIDIA L4 GPU with 24Gb VRAM.
I located the source of this error message in https://github.com/ollama/ollama/blob/main/runner/ollamarunner/runner.go but I do not see how to circumvent this behavior.
Is this a bug, or a normal behavior that can be circumvented by modifying some parameters?

Relevant log output

OS

Linux

GPU

Nvidia

CPU

AMD

Ollama version

0.18.0

Originally created by @fchahun on GitHub (Mar 30, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/15147 ### What is the issue? I am facing a problem with qwen3.5:27b when submitting a list of 2-3 images (representing the pages of a same document) along with a single text prompt requiring content extraction of the images. With qwen3.5:397b-cloud, everything runs fine as expected. But with qwen3.5:27b, using exactly the same prompt and same images, I consistently get the following error message: "**Internal Server Error -- Failed to create new sequence: SameBatch may not be specified within numKeep (index: 3 numKeep: 4 SameBatch: 2017)"**. This is true whether using the "generate" or "chat" API endpoints. I am using Ollama 0.18.0 running on an NVIDIA L4 GPU with 24Gb VRAM. I located the source of this error message in https://github.com/ollama/ollama/blob/main/runner/ollamarunner/runner.go but I do not see how to circumvent this behavior. Is this a bug, or a normal behavior that can be circumvented by modifying some parameters? ### Relevant log output ```shell ``` ### OS Linux ### GPU Nvidia ### CPU AMD ### Ollama version 0.18.0
GiteaMirror added the bug label 2026-05-05 02:27:22 -05:00
GiteaMirror commented 2026-05-05 02:27:23 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Mar 30, 2026):

Have you tried increasing the context window?

Server logs will aid in debugging.

<!-- gh-comment-id:4156841879 --> @rick-github commented on GitHub (Mar 30, 2026): Have you tried increasing the [context window](https://github.com/ollama/ollama/blob/main/docs/faq.mdx#how-can-i-specify-the-context-window-size)? [Server logs](https://docs.ollama.com/troubleshooting) will aid in debugging.
GiteaMirror commented 2026-05-05 02:27:24 -05:00
Author
Owner
Copy Link

@fchahun commented on GitHub (Mar 31, 2026):

Thank you: increasing num_ctx to 16384 did solve the problem (actually I found that, for this run, 12750 tokens at least are required based on reported prompt + eval count)

Two things actually fooled me in this case:

  • The Ollama doc says that default context windows is 32K for 24-48 GiB VRAM but it happens that the "24 GiB VRAM" NVIDIA L4 GPU is seen with slightly less than 24GiB VRAM, hence a default num_ctx set at 2K.
  • Qwen3.5:27b is very verbose in thinking output, even with think="low", which adds up to token count

But would it be possible for such cases to consider having a more explicit error message, hinting towards increasing the context window, rather than simply "Failed to create new sequence: SameBatch may not be specified within numKeep" ?

<!-- gh-comment-id:4163239327 --> @fchahun commented on GitHub (Mar 31, 2026): Thank you: increasing **num_ctx** to 16384 did solve the problem (actually I found that, for this run, 12750 tokens at least are required based on reported prompt + eval count) Two things actually fooled me in this case: - The Ollama doc says that default context windows is **32K for 24-48 GiB VRAM** but it happens that the "24 GiB VRAM" NVIDIA L4 GPU is seen with slightly less than 24GiB VRAM, hence a default **num_ctx** set at 2K. - Qwen3.5:27b is **very** verbose in thinking output, even with think="low", which adds up to token count But would it be possible for such cases to consider having a **more explicit error message**, hinting towards increasing the context window, rather than simply "_Failed to create new sequence: SameBatch may not be specified within numKeep_" ?
GiteaMirror commented 2026-05-05 02:27:24 -05:00
Author
Owner
Copy Link

@PureBlissAK commented on GitHub (Apr 18, 2026):

🤖 Automated Triage & Analysis Report

Issue: #15147
Analyzed: 2026-04-18T18:23:00.237216

Analysis

  • Type: unknown
  • Severity: medium
  • Components: unknown

Implementation Plan

  • Effort: medium
  • Steps:

This issue has been triaged and marked for implementation.

<!-- gh-comment-id:4274311062 --> @PureBlissAK commented on GitHub (Apr 18, 2026): <!-- ollama-issue-orchestrator:v1 issue:15147 --> ## 🤖 Automated Triage & Analysis Report **Issue**: #15147 **Analyzed**: 2026-04-18T18:23:00.237216 ### Analysis - **Type**: unknown - **Severity**: medium - **Components**: unknown ### Implementation Plan - **Effort**: medium - **Steps**: *This issue has been triaged and marked for implementation.*
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#71757
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?