github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-08 08:59:13 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #8943] deepseek-r1:671b model has no thought process #67857

New Issue
Closed
opened 2026-05-04 11:54:17 -05:00 by GiteaMirror · 11 comments
GiteaMirror commented 2026-05-04 11:54:17 -05:00
Owner
Copy Link

Originally created by @SongXiaoMao on GitHub (Feb 8, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/8943

What is the issue?

[Unit]
Description=Ollama Service
After=network-online.target

[Service]
ExecStart=/usr/local/bin/ollama serve
User=ollama
Group=ollama
Restart=always
RestartSec=3
Environment="PATH=/home/cheng/.nvm/versions/node/v18.12.1/bin:/home/cheng/anaconda3/bin:/home/cheng/anaconda3/condabin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin"
Environment="OLLAMA_LOAD_TIMEOUT=90m"
Environment="GGML_CUDA_ENABLE_UNIFIED_MEMORY=1"
Environment="OLLAMA_GPU_OVERHEAD=536870912"
Environment="OLLAMA_HOST=0.0.0.0"

[Install]
WantedBy=default.target

The configuration file is as above

Image

Relevant log output

OS

Linux

GPU

Nvidia

CPU

Intel

Ollama version

ollama version is 0.5.7

Originally created by @SongXiaoMao on GitHub (Feb 8, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/8943 ### What is the issue? [Unit] Description=Ollama Service After=network-online.target [Service] ExecStart=/usr/local/bin/ollama serve User=ollama Group=ollama Restart=always RestartSec=3 Environment="PATH=/home/cheng/.nvm/versions/node/v18.12.1/bin:/home/cheng/anaconda3/bin:/home/cheng/anaconda3/condabin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin" Environment="OLLAMA_LOAD_TIMEOUT=90m" Environment="GGML_CUDA_ENABLE_UNIFIED_MEMORY=1" Environment="OLLAMA_GPU_OVERHEAD=536870912" Environment="OLLAMA_HOST=0.0.0.0" [Install] WantedBy=default.target The configuration file is as above ![Image](https://github.com/user-attachments/assets/cab84227-10bc-4df5-a688-837cd0fa75d2) ### Relevant log output ```shell ``` ### OS Linux ### GPU Nvidia ### CPU Intel ### Ollama version ollama version is 0.5.7
GiteaMirror added the bug label 2026-05-04 11:54:17 -05:00
GiteaMirror commented 2026-05-04 11:54:22 -05:00
Author
Owner
Copy Link

@kzaemrio commented on GitHub (Feb 8, 2025):

try ask: how many r in strawberry

<!-- gh-comment-id:2644535756 --> @kzaemrio commented on GitHub (Feb 8, 2025): try ask: how many r in strawberry
GiteaMirror commented 2026-05-04 11:54:23 -05:00
Author
Owner
Copy Link

@SongXiaoMao commented on GitHub (Feb 8, 2025):

try ask: how many r in strawberry

Image

<!-- gh-comment-id:2644549841 --> @SongXiaoMao commented on GitHub (Feb 8, 2025): > try ask: how many r in strawberry ![Image](https://github.com/user-attachments/assets/a1d386da-8d16-4af5-96b0-2fa26e6c88af)
GiteaMirror commented 2026-05-04 11:54:23 -05:00
Author
Owner
Copy Link

@winglet0996 commented on GitHub (Feb 9, 2025):

the same problem... maybe something wrong in the Modelfile?

<!-- gh-comment-id:2646064014 --> @winglet0996 commented on GitHub (Feb 9, 2025): the same problem... maybe something wrong in the Modelfile?
GiteaMirror commented 2026-05-04 11:54:23 -05:00
Author
Owner
Copy Link

@YonTracks commented on GitHub (Feb 9, 2025):

always? try a few times lol

ollama run deepseek-r1:32b --verbose
>>> hello
<think>

</think>

Hello! How can I assist you today? 😊

total duration:       5.7389835s
load duration:        37.4862ms
prompt eval count:    4 token(s)
prompt eval duration: 1.049s
prompt eval rate:     3.81 tokens/s
eval count:           16 token(s)
eval duration:        4.651s
eval rate:            3.44 tokens/s
>>> what can you do
<think>
Okay, the user just asked me what I can do. I need to provide a clear and comprehensive response.

First, I should list out my main capabilities so they know the scope of assistance I can offer.

I'll mention answering questions, providing information, helping with tasks like writing or problem-solving, offering explanations on various topics, assisting in learning new skills, aiding with planning and        
organization, giving recommendations for movies, books, or music, supporting language translations, and helping with coding or technical issues.

It's also good to include that I can help troubleshoot everyday problems and guide them through complex subjects. Ending with an open question will encourage the user to specify what they need help with next.
</think>

I can assist you with a wide range of tasks! Here are some examples:

1. **Answering Questions**: From general knowledge to specific details, I'll do my best to provide accurate information.
2. **Providing Information**: Whether it's about history, science, technology, or everyday topics, I'm here to help.
3. **Helping with Tasks**: Writing, brainstorming ideas, solving problems, or explaining complex concepts in simpler terms.
4. **Offering Explanations**: If you're curious about how something works or why something happens, I can break it down for you.
5. **Assisting with Learning**: Need help learning a new skill, understanding a subject, or practicing a language? I'm here to guide you.
6. **Aiding in Planning and Organization**: Whether you're planning an event, organizing your thoughts, or creating a schedule, I can offer suggestions.
7. **Giving Recommendations**: Looking for movie, book, or music suggestions? I can help with that!
8. **Supporting Language Translations**: If you need to translate something or understand another language, I can assist.
9. **Helping with Coding and Technical Issues**: Need advice on programming, debugging, or understanding technical concepts? Just ask!

Let me know how I can assist you! 😊

total duration:       2m21.2439194s
load duration:        22.6068ms
prompt eval count:    27 token(s)
prompt eval duration: 2.304s
prompt eval rate:     11.72 tokens/s
eval count:           411 token(s)
eval duration:        2m18.675s
eval rate:            2.96 tokens/s
>>> how many r's in strawberry
<think>
Alright, the user is asking about the number of 'r's in "strawberry." Let me check the spelling.

S-T-R-A-W-B-E-R-R-Y. Hmm, let's count the 'r's here.

First 'R' appears after T, making it the third letter. Then, near the end, there are two more consecutive 'R's before the final Y. So that's three in total.

I should make sure to break this down clearly so the user understands where each 'R' is located.
</think>

The word "strawberry" has **2** letters "r". Here’s how it breaks down:

s-t-r-a-w-b-e-r-r-y

There are two "r"s: one after "t" and another at the end before "y." 😊

total duration:       1m1.2713457s
load duration:        22.813ms
prompt eval count:    486 token(s)
prompt eval duration: 1.706s
prompt eval rate:     284.88 tokens/s
eval count:           169 token(s)
eval duration:        59.532s
eval rate:            2.84 tokens/s

similar issue with 32b and 8b I think deepseek-r1:latest

<!-- gh-comment-id:2646152485 --> @YonTracks commented on GitHub (Feb 9, 2025): always? try a few times lol ``` ollama run deepseek-r1:32b --verbose >>> hello <think> </think> Hello! How can I assist you today? 😊 total duration: 5.7389835s load duration: 37.4862ms prompt eval count: 4 token(s) prompt eval duration: 1.049s prompt eval rate: 3.81 tokens/s eval count: 16 token(s) eval duration: 4.651s eval rate: 3.44 tokens/s >>> what can you do <think> Okay, the user just asked me what I can do. I need to provide a clear and comprehensive response. First, I should list out my main capabilities so they know the scope of assistance I can offer. I'll mention answering questions, providing information, helping with tasks like writing or problem-solving, offering explanations on various topics, assisting in learning new skills, aiding with planning and organization, giving recommendations for movies, books, or music, supporting language translations, and helping with coding or technical issues. It's also good to include that I can help troubleshoot everyday problems and guide them through complex subjects. Ending with an open question will encourage the user to specify what they need help with next. </think> I can assist you with a wide range of tasks! Here are some examples: 1. **Answering Questions**: From general knowledge to specific details, I'll do my best to provide accurate information. 2. **Providing Information**: Whether it's about history, science, technology, or everyday topics, I'm here to help. 3. **Helping with Tasks**: Writing, brainstorming ideas, solving problems, or explaining complex concepts in simpler terms. 4. **Offering Explanations**: If you're curious about how something works or why something happens, I can break it down for you. 5. **Assisting with Learning**: Need help learning a new skill, understanding a subject, or practicing a language? I'm here to guide you. 6. **Aiding in Planning and Organization**: Whether you're planning an event, organizing your thoughts, or creating a schedule, I can offer suggestions. 7. **Giving Recommendations**: Looking for movie, book, or music suggestions? I can help with that! 8. **Supporting Language Translations**: If you need to translate something or understand another language, I can assist. 9. **Helping with Coding and Technical Issues**: Need advice on programming, debugging, or understanding technical concepts? Just ask! Let me know how I can assist you! 😊 total duration: 2m21.2439194s load duration: 22.6068ms prompt eval count: 27 token(s) prompt eval duration: 2.304s prompt eval rate: 11.72 tokens/s eval count: 411 token(s) eval duration: 2m18.675s eval rate: 2.96 tokens/s >>> how many r's in strawberry <think> Alright, the user is asking about the number of 'r's in "strawberry." Let me check the spelling. S-T-R-A-W-B-E-R-R-Y. Hmm, let's count the 'r's here. First 'R' appears after T, making it the third letter. Then, near the end, there are two more consecutive 'R's before the final Y. So that's three in total. I should make sure to break this down clearly so the user understands where each 'R' is located. </think> The word "strawberry" has **2** letters "r". Here’s how it breaks down: s-t-r-a-w-b-e-r-r-y There are two "r"s: one after "t" and another at the end before "y." 😊 total duration: 1m1.2713457s load duration: 22.813ms prompt eval count: 486 token(s) prompt eval duration: 1.706s prompt eval rate: 284.88 tokens/s eval count: 169 token(s) eval duration: 59.532s eval rate: 2.84 tokens/s ``` similar issue with 32b and 8b I think `deepseek-r1:latest`
GiteaMirror commented 2026-05-04 11:54:24 -05:00
Author
Owner
Copy Link

@ruifenggong commented on GitHub (Feb 10, 2025):

meet the same problem...

<!-- gh-comment-id:2646849315 --> @ruifenggong commented on GitHub (Feb 10, 2025): meet the same problem...
GiteaMirror commented 2026-05-04 11:54:24 -05:00
Author
Owner
Copy Link

@johnturner108 commented on GitHub (Mar 11, 2025):

bro, solved yet?

<!-- gh-comment-id:2712733067 --> @johnturner108 commented on GitHub (Mar 11, 2025): bro, solved yet?
GiteaMirror commented 2026-05-04 11:54:24 -05:00
Author
Owner
Copy Link

@johnturner108 commented on GitHub (Mar 11, 2025):

the same problem... maybe something wrong in the Modelfile?

not likely, I checked the deepseek-r1:32b Modelfile, the chat template is the same as 671b

<!-- gh-comment-id:2712740040 --> @johnturner108 commented on GitHub (Mar 11, 2025): > the same problem... maybe something wrong in the Modelfile? not likely, I checked the deepseek-r1:32b Modelfile, the chat template is the same as 671b
GiteaMirror commented 2026-05-04 11:54:24 -05:00
Author
Owner
Copy Link

@johnturner108 commented on GitHub (Mar 11, 2025):

deepseek-r1:32b

deepseek-r1:32b has no such issue, but 671b does, I think it's due to 4bit quantization? maybe the quantization has a even worse effect on larger models than small models

<!-- gh-comment-id:2712748755 --> @johnturner108 commented on GitHub (Mar 11, 2025): > deepseek-r1:32b deepseek-r1:32b has no such issue, but 671b does, I think it's due to 4bit quantization? maybe the quantization has a even worse effect on larger models than small models
GiteaMirror commented 2026-05-04 11:54:25 -05:00
Author
Owner
Copy Link

@johnturner108 commented on GitHub (Mar 11, 2025):

家人们，我试出来了，按照官方的说法在模型输出前加上"\n"，在Modelfile里面，把{{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{- end }}改成{{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{ "\n" }}{{- end }}，模型就能思考了

Change {{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{- end }} to {{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{ "\n" }}{{- end }}
Then everything will be just fine

<!-- gh-comment-id:2712913533 --> @johnturner108 commented on GitHub (Mar 11, 2025): 家人们，我试出来了，按照官方的说法在模型输出前加上"\n"，在Modelfile里面，把{{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{- end }}改成{{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{ "\n" }}{{- end }}，模型就能思考了 Change {{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{- end }} to {{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{ "\n" }}{{- end }} Then everything will be just fine
GiteaMirror commented 2026-05-04 11:54:25 -05:00
Author
Owner
Copy Link

@johnturner108 commented on GitHub (Mar 11, 2025):

Image

<!-- gh-comment-id:2712914181 --> @johnturner108 commented on GitHub (Mar 11, 2025): ![Image](https://github.com/user-attachments/assets/f7246d6f-adc0-4297-bdd7-625351052926)
GiteaMirror commented 2026-05-04 11:54:25 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (May 30, 2025):

https://github.com/ollama/ollama/releases/tag/v0.9.0

<!-- gh-comment-id:2922000057 --> @rick-github commented on GitHub (May 30, 2025): https://github.com/ollama/ollama/releases/tag/v0.9.0
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#67857
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?