ollama - 💡(How to fix) Fix Gemma4 26B and 31B miss some UTF-8 characters in polish [4 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
ollama/ollama#15240Fetched 2026-04-08 02:33:50
View on GitHub
Comments
4
Participants
2
Timeline
6
Reactions
0
Author
Participants
Timeline (top)
commented ×4closed ×1labeled ×1

Code Example

PowerShell 7.6.0
PS C:\Users\lukas> ollama run gemma4:31b
>>> cześć, proszę zidentyfikuj się
Thinking...
"cześ, prosz zidentyfikuj si" (Hi, please identify yourself).
Polish.
The user wants to know who/what I am.

    *   I am a large language model.
    *   I was trained by Google.
    *   I don't have a personal name, feelings, or a physical body.
    *   My purpose is to help with a variety of tasks (writing, translating, coding, answering questions, etc.).

    *   *Option 1 (Simple):* Cześ! Jestem modelem jzykowym wytrenowanym przez Google. (Hi! I am a language model
trained by Google.)
    *   *Option 2 (Detailed):* Cześ! Jestem dużym modelem jzykowym, stworzonym przez Google. Mog pomaga w pisaniu
tekst, tzeniu, programowaniu i odpowiadaniu na pytania. (Hi! I am a large language model, created by Google. I can
help with writing texts, translating, programming, and answering questions.)
    *   *Option 3 (Conversational):* Hej! Jestem sztuczn inteligencj od Google. W czym mog Ci dzisiaj pom? (Hey! I
am an AI from Google. How can I help you today?)

    *   A combination of being clear, professional, and helpful is best.

    *   *Draft:* "Cześ! Jestem dużym modelem jzykowym, wytrenowanym przez Google." (Hi! I am a large language
model, trained by Google.)
...done thinking.

Cześ! Jestem dużym modelem jzykowym, wytrenowanym przez Google.

W skr: jestem sztuczn inteligencj, kt potrafi przetwarza i generowa tekst w wielu jzykach. Mog Ci pom w wielu
rzeczach, na przyk:

* **Odpowiadaniu na pytania** z r dziedzin.
* **Pisaniu tekst** (maile, wypracowania, wiersze, kody programistyczne).
* **Tumaczeniu** treści midzy jzykami.
* **Podsumowywaniu** dich artyku.
* **Planowaniu** podr, nauki czy codziennych zadań.

W czym mog Ci dzisiaj pom?
PS C:\Users\lukas> ollama --version
ollama version is 0.20.0-rc0
RAW_BUFFERClick to expand / collapse

What is the issue?

In model responses I see that some characters are missing, even when model is repeating verbatim my own prompt. This issue does not appear on Vertex AI and in Google AI Studio, so it's unlikely to be model problem. Due to missing both UTF and latin characters following are missing, I suspect this is a multibyte encoding issue.

For example

* *Option 3 (Conversational):* Hej! Jestem sztuczn inteligencj od Google. W czym mog Ci dzisiaj pom? (Hey! I am an AI from Google. How can I help you today?)

should be

* *Option 3 (Conversational):* Hej! Jestem sztuczną inteligencją od Google. W czym mogę Ci dzisiaj pomóc? (Hey! I am an AI from Google. How can I help you today?)

Relevant log output

PowerShell 7.6.0
PS C:\Users\lukas> ollama run gemma4:31b
>>> cześć, proszę zidentyfikuj się
Thinking...
"cześ, prosz zidentyfikuj si" (Hi, please identify yourself).
Polish.
The user wants to know who/what I am.

    *   I am a large language model.
    *   I was trained by Google.
    *   I don't have a personal name, feelings, or a physical body.
    *   My purpose is to help with a variety of tasks (writing, translating, coding, answering questions, etc.).

    *   *Option 1 (Simple):* Cześ! Jestem modelem jzykowym wytrenowanym przez Google. (Hi! I am a language model
trained by Google.)
    *   *Option 2 (Detailed):* Cześ! Jestem dużym modelem jzykowym, stworzonym przez Google. Mog pomaga w pisaniu
tekst, tzeniu, programowaniu i odpowiadaniu na pytania. (Hi! I am a large language model, created by Google. I can
help with writing texts, translating, programming, and answering questions.)
    *   *Option 3 (Conversational):* Hej! Jestem sztuczn inteligencj od Google. W czym mog Ci dzisiaj pom? (Hey! I
am an AI from Google. How can I help you today?)

    *   A combination of being clear, professional, and helpful is best.

    *   *Draft:* "Cześ! Jestem dużym modelem jzykowym, wytrenowanym przez Google." (Hi! I am a large language
model, trained by Google.)
...done thinking.

Cześ! Jestem dużym modelem jzykowym, wytrenowanym przez Google.

W skr: jestem sztuczn inteligencj, kt potrafi przetwarza i generowa tekst w wielu jzykach. Mog Ci pom w wielu
rzeczach, na przyk:

* **Odpowiadaniu na pytania** z r dziedzin.
* **Pisaniu tekst** (maile, wypracowania, wiersze, kody programistyczne).
* **Tumaczeniu** treści midzy jzykami.
* **Podsumowywaniu** dich artyku.
* **Planowaniu** podr, nauki czy codziennych zadań.

W czym mog Ci dzisiaj pom?
PS C:\Users\lukas> ollama --version
ollama version is 0.20.0-rc0

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

ollama version is 0.20.0-rc0

extent analysis

TL;DR

The issue can likely be resolved by adjusting the character encoding settings in the ollama configuration to properly handle multibyte characters.

Guidance

  • Verify that the ollama configuration is set to use a character encoding that supports multibyte characters, such as UTF-8.
  • Check the documentation for ollama version 0.20.0-rc0 to see if there are any known issues or configuration options related to character encoding.
  • Test the ollama model with a different input prompt that contains multibyte characters to see if the issue is specific to certain characters or prompts.
  • Consider updating to a newer version of ollama if one is available, as the issue may have been resolved in a later release.

Example

No code snippet is provided as the issue is related to character encoding and configuration, rather than code.

Notes

The issue may be specific to the Windows operating system or the ollama version being used. Further testing and debugging may be necessary to determine the root cause of the issue.

Recommendation

Apply workaround: Adjust the character encoding settings in the ollama configuration to properly handle multibyte characters, as this is the most likely cause of the issue.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

ollama - 💡(How to fix) Fix Gemma4 26B and 31B miss some UTF-8 characters in polish [4 comments, 2 participants]