Disable tensor creation logging

Oct 13, 2025 by windailys.com 72 views

Huge server.log when using gemma3 and OLLAMA_DEBUG=1

Introduction

When using the gemma3 model with the OLLAMA_DEBUG=1 flag, the server.log file can become extremely large, reaching sizes of up to 4 MB. This issue is not present when using the llama3.2 model. In this article, we will investigate the cause of this problem and provide potential solutions.

What is the issue?

When using gemma3 with OLLAMA_DEBUG=1, the debug-log in server.log is huge, and the generated request does not include system-prompt like on other models. This issue is specific to the gemma3 model and is not present when using the llama3.2 model.

Relevant log output

The relevant log output from the server.log file is as follows:

time=2025-03-12T18:33:39.415+01:00 level=DEBUG source=ggml.go:220 msg="created tensor" name=v.blk.18.mlp.fc2.bias shape=[1152] dtype=0 buffer_type=CUDA0
time=2025-03-12T18:33:39.415+01:00 level=DEBUG source=ggml.go:220 msg="created tensor" name=v.blk.18.mlp.fc2.weight shape="[4304 1152]" dtype=1 buffer_type=CUDA0
time=2025-03-12T18:33:39.415+01:00 level=DEBUG source=ggml.go:220 msg="created tensor" name=v.blk.19.attn_k.bias shape=[1152] dtype=0 buffer_type=CUDA0
time=2025-03-12T18:33:39.415+01:00 level=DEBUG source=ggml.go:220 msg="created tensor" name=v.blk.19.attn_k.weight shape="[1152 1152]" dtype=1 buffer_type=CUDA0
...
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:-1 n:5 runes:[9601 100 97 110 107]}" right="{p:0 n:5 runes:[]}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:2 n:4 runes:[]}" right="{p:2 n:5 runes:[]}"
...

OS, GPU, and CPU

The issue is being reported on a Windows system with an Nvidia GPU and an Intel CPU.

Ollama version

The Ollama version being used is 0.6.0.

Potential solutions

Based on the log output, it appears that the issue is related to the creation of tensors and the processing of text. Here are some potential solutions:

Reduce the log level: Try reducing the log level to see if it reduces the size of the server.log file.
Disable tensor creation logging: Try disabling the logging of tensor creation to see if it reduces the size of the server.log file.
Optimize the text processing: Try optimizing the text processing code to reduce the number of log messages generated.
Upgrade to a newer Ollama version: Try upgrading to a newer version of Ollama to see if the issue is fixed.

Conclusion

The huge server.log file when using gemma3 and OLLAMA_DEBUG=1 is a complex issue that requires further investigation. Based on the log output, it appears that the issue is related to the creation of tensors and the processing of text. Potential solutions include reducing the log level, disabling tensor creation logging, optimizing the text processing code, and upgrading to a newer Ollama version.

Recommendations

If you are experiencing this issue, we recommend trying the following:

Reduce the log level: Try reducing the log level to see if it reduces the size of the server.log file.
Disable tensor creation logging: Try disabling the logging of tensor creation to see if it reduces the size of the server.log file.
Optimize the text processing: Try optimizing the text processing code to reduce the number of log messages generated.
Upgrade to a newer Ollama version: Try upgrading to a newer version of Ollama to see if the issue is fixed.

Future work

We plan to investigate this issue further and provide a more detailed analysis of the log output. We will also work on optimizing the text processing code to reduce the number of log messages generated.

Acknowledgments

We would like to thank the Ollama community for their feedback and suggestions on this issue. We are grateful for their help in resolving this issue.
Huge server.log when using gemma3 and OLLAMA_DEBUG=1: Q&A

Q: What is the issue with the server.log file when using gemma3 and OLLAMA_DEBUG=1?

A: The server.log file becomes extremely large, reaching sizes of up to 4 MB, when using gemma3 with OLLAMA_DEBUG=1. This issue is not present when using the llama3.2 model.

Q: What is the cause of this issue?

A: The cause of this issue is related to the creation of tensors and the processing of text. The log output shows a large number of log messages generated during the creation of tensors and the processing of text.

Q: How can I reduce the size of the server.log file?

A: There are several potential solutions to reduce the size of the server.log file:

Reduce the log level: Try reducing the log level to see if it reduces the size of the server.log file.
Disable tensor creation logging: Try disabling the logging of tensor creation to see if it reduces the size of the server.log file.
Optimize the text processing: Try optimizing the text processing code to reduce the number of log messages generated.
Upgrade to a newer Ollama version: Try upgrading to a newer version of Ollama to see if the issue is fixed.

Q: What are the potential solutions to this issue?

A: The potential solutions to this issue include:

Reduce the log level: Try reducing the log level to see if it reduces the size of the server.log file.
Disable tensor creation logging: Try disabling the logging of tensor creation to see if it reduces the size of the server.log file.
Optimize the text processing: Try optimizing the text processing code to reduce the number of log messages generated.
Upgrade to a newer Ollama version: Try upgrading to a newer version of Ollama to see if the issue is fixed.

Q: How can I optimize the text processing code?

A: To optimize the text processing code, you can try the following:

Reduce the number of log messages: Try reducing the number of log messages generated during the text processing.
Use a more efficient algorithm: Try using a more efficient algorithm for text processing.
Use a caching mechanism: Try using a caching mechanism to reduce the number of log messages generated.

Q: What are the benefits of upgrading to a newer Ollama version?

A: Upgrading to a newer Ollama version may fix the issue with the server.log file and provide other benefits, such as:

Improved performance: Newer versions of Ollama may provide improved performance.
New features: Newer versions of Ollama may provide new features.
Bug fixes: Newer versions of Ollama may fix bugs.

Q: How can I upgrade to a newer Ollama version?

A: To upgrade to a newer Ollama version, you can try the following:

Check the Ollama website: Check the Ollama website for the latest version of Ollama.
Download the new version: Download the new version of Ollama.
Install the new version: Install the new version of Ollama.

Q: What are the potential risks of upgrading to a newer Ollama version?

A: Upgrading to a newer Ollama version may have potential risks, such as:

Compatibility issues: The new version of Ollama may not be compatible with your existing code.
Performance issues: The new version of Ollama may have performance issues.
Bug fixes: The new version of Ollama may introduce new bugs.

Q: How can I mitigate the potential risks of upgrading to a newer Ollama version?

A: To mitigate the potential risks of upgrading to a newer Ollama version, you can try the following:

Test the new version: Test the new version of Ollama before upgrading.
Check the documentation: Check the documentation for the new version of Ollama.
Seek support: Seek support from the Ollama community or a professional.