github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #9698] Huge server.log when using gemma3 and OLLAMA_DEBUG=1 #32092

New Issue
Closed
opened 2026-04-22 13:01:06 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-22 13:01:06 -05:00
Owner
Copy Link

Originally created by @TheMasterFX on GitHub (Mar 12, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/9698

What is the issue?

When using gemma3 with OLLAMA_DEBUG=1 the debug-log in server.log is huge (like 4 MB after some chatting with some images). Also I do not see the generated request including system-prompt like on other models.
When using e.g. llama3.2 everthing is fine.
I reduced the log output just to show what I see

Relevant log output

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=2025-03-12T18:33:39.415+01:00 level=DEBUG source=ggml.go:220 msg="created tensor" name=v.blk.19.attn_output.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_output.weight shape="[1152 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_q.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_q.weight shape="[1152 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_v.bias shape=[1152] dtype=0 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:[]}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:192 msg=merges merges="[{p:-1 n:5 runes:[9601 100 97 110 107]} {p:0 n:2 runes:[]} {p:0 n:5 runes:[]} {p:2 n:4 runes:[]} {p:2 n:5 runes:[]} {p:0 n:6 runes:[101]}]"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:132 msg=tokenizer merges="[{p:-1 n:1 runes:[9601]} {p:0 n:2 runes:[109]} {p:1 n:3 runes:[97]} {p:2 n:4 runes:[99]} {p:3 n:5 runes:[104]} {p:4 n:6 runes:[115]} {p:5 n:7 runes:[116]}]"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:0 b:1 score:-26}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:5 b:6 score:-46}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:3 b:4 score:-80}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:2 b:3 score:-58}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:1 b:2 score:-347}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:4 b:5 score:-26321}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:-1 n:1 runes:[9601]}" right="{p:0 n:2 runes:[109]}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:4 n:6 runes:[115]}" right="{p:5 n:7 runes:[116]}"
time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:0 n:3 runes:[97]}" right="{p:2 n:4 runes:[99]}"
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:[104]}"
...
time=2025-03-12T18:33:41.481+01:00 level=DEBUG source=process_text_spm.go:212 msg="adding bos token to prompt" id=2
time=2025-03-12T18:33:41.481+01:00 level=DEBUG source=cache.go:134 msg="loading cache slot" id=0 cache=0 prompt=160 used=0 remaining=160
time=2025-03-12T18:33:41.721+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[19058] text=Okay
time=2025-03-12T18:33:41.744+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236764] text=,
time=2025-03-12T18:33:41.761+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[1531] text=" let"
time=2025-03-12T18:33:41.776+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236858] text=’
time=2025-03-12T18:33:41.789+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236751] text=s
time=2025-03-12T18:33:41.803+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[78543] text=" brainstorm"
time=2025-03-12T18:33:41.817+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[1070] text=" some"
time=2025-03-12T18:33:41.830+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[6549] text=" ideas"
time=2025-03-12T18:33:41.844+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[573] text=" for"
time=2025-03-12T18:33:41.856+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[822] text=" your"
time=2025-03-12T18:33:41.869+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[9137] text=" evening"
time=2025-03-12T18:33:41.883+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236888] text=!

OS

Windows

GPU

Nvidia

CPU

Intel

Ollama version

0.6.0

Originally created by @TheMasterFX on GitHub (Mar 12, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/9698 ### What is the issue? When using gemma3 with OLLAMA_DEBUG=1 the debug-log in server.log is huge (like 4 MB after some chatting with some images). Also I do not see the generated request including system-prompt like on other models. When using e.g. llama3.2 everthing is fine. I reduced the log output just to show what I see ### Relevant log output ```shell 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=2025-03-12T18:33:39.415+01:00 level=DEBUG source=ggml.go:220 msg="created tensor" name=v.blk.19.attn_output.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_output.weight shape="[1152 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_q.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_q.weight shape="[1152 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_v.bias shape=[1152] dtype=0 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:[]}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:192 msg=merges merges="[{p:-1 n:5 runes:[9601 100 97 110 107]} {p:0 n:2 runes:[]} {p:0 n:5 runes:[]} {p:2 n:4 runes:[]} {p:2 n:5 runes:[]} {p:0 n:6 runes:[101]}]" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:132 msg=tokenizer merges="[{p:-1 n:1 runes:[9601]} {p:0 n:2 runes:[109]} {p:1 n:3 runes:[97]} {p:2 n:4 runes:[99]} {p:3 n:5 runes:[104]} {p:4 n:6 runes:[115]} {p:5 n:7 runes:[116]}]" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:0 b:1 score:-26}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:5 b:6 score:-46}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:3 b:4 score:-80}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:2 b:3 score:-58}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:1 b:2 score:-347}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:159 msg=candidate candidate="&{a:4 b:5 score:-26321}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:-1 n:1 runes:[9601]}" right="{p:0 n:2 runes:[109]}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:4 n:6 runes:[115]}" right="{p:5 n:7 runes:[116]}" time=2025-03-12T18:33:41.445+01:00 level=DEBUG source=process_text_spm.go:167 msg=pair left="{p:0 n:3 runes:[97]}" right="{p:2 n:4 runes:[99]}" 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:[104]}" ... time=2025-03-12T18:33:41.481+01:00 level=DEBUG source=process_text_spm.go:212 msg="adding bos token to prompt" id=2 time=2025-03-12T18:33:41.481+01:00 level=DEBUG source=cache.go:134 msg="loading cache slot" id=0 cache=0 prompt=160 used=0 remaining=160 time=2025-03-12T18:33:41.721+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[19058] text=Okay time=2025-03-12T18:33:41.744+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236764] text=, time=2025-03-12T18:33:41.761+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[1531] text=" let" time=2025-03-12T18:33:41.776+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236858] text=’ time=2025-03-12T18:33:41.789+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236751] text=s time=2025-03-12T18:33:41.803+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[78543] text=" brainstorm" time=2025-03-12T18:33:41.817+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[1070] text=" some" time=2025-03-12T18:33:41.830+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[6549] text=" ideas" time=2025-03-12T18:33:41.844+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[573] text=" for" time=2025-03-12T18:33:41.856+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[822] text=" your" time=2025-03-12T18:33:41.869+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[9137] text=" evening" time=2025-03-12T18:33:41.883+01:00 level=DEBUG source=process_text_spm.go:244 msg=decoded ids=[236888] text=! ``` ### OS Windows ### GPU Nvidia ### CPU Intel ### Ollama version 0.6.0
GiteaMirror added the bug label 2026-04-22 13:01:06 -05:00
GiteaMirror commented 2026-04-22 13:01:09 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 12, 2025):

Yes, this is expected as gemma3 is running on the new engine w/ a new tokenizer and we needed a lot of debugging information. We'll make this less verbose in an upcoming release (most of the tokenizer debugging can probably go to a new Trace log level)

<!-- gh-comment-id:2718832343 --> @pdevine commented on GitHub (Mar 12, 2025): Yes, this is expected as `gemma3` is running on the new engine w/ a new tokenizer and we needed a lot of debugging information. We'll make this less verbose in an upcoming release (most of the tokenizer debugging can probably go to a new Trace log level)
GiteaMirror commented 2026-04-22 13:01:10 -05:00
Author
Owner
Copy Link

@TheMasterFX commented on GitHub (Mar 12, 2025):

Yes, this is expected as gemma3 is running on the new engine w/ a new tokenizer and we needed a lot of debugging information. We'll make this less verbose in an upcoming release (most of the tokenizer debugging can probably go to a new Trace log level)

That's almost what I thought. Maybe introducing different levels of verbosity makes sense?

<!-- gh-comment-id:2718850527 --> @TheMasterFX commented on GitHub (Mar 12, 2025): > Yes, this is expected as `gemma3` is running on the new engine w/ a new tokenizer and we needed a lot of debugging information. We'll make this less verbose in an upcoming release (most of the tokenizer debugging can probably go to a new Trace log level) That's almost what I thought. Maybe introducing different levels of verbosity makes sense?
GiteaMirror commented 2026-04-22 13:01:12 -05:00
Author
Owner
Copy Link

@dhiltgen commented on GitHub (Jul 5, 2025):

Unless you're trying to troubleshoot a problem, you shouldn't set the debug log level. The volume of logging in debug will fluctuate over time as we add/remove logging calls in areas of the system that are new or maturing.

<!-- gh-comment-id:3039827226 --> @dhiltgen commented on GitHub (Jul 5, 2025): Unless you're trying to troubleshoot a problem, you shouldn't set the debug log level. The volume of logging in debug will fluctuate over time as we add/remove logging calls in areas of the system that are new or maturing.
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#32092
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?