Is your feature request related to a problem? Please describe.
Since reasoning models are now mainstream, we are now often facing an issue of too many reasoning tokens thrown right in our face. This can flood the UI and make your laptop levitating because of coolers blasting at 100% due to UI update cycle. But the issue here is mostly about UI clutter than performance.
Describe the solution you'd like
I'd like to have a foldable reasoning tokens widget in OpenWebUI, much like in other UIs (see below)
Originally created by @0x7CFE on GitHub (Jan 22, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/8769
# Feature Request
**Is your feature request related to a problem? Please describe.**
Since reasoning models are now mainstream, we are now often facing an issue of too many reasoning tokens thrown right in our face. This can flood the UI and make your laptop levitating because of coolers blasting at 100% due to UI update cycle. But the issue here is mostly about UI clutter than performance.
**Describe the solution you'd like**
I'd like to have a foldable reasoning tokens widget in OpenWebUI, much like in other UIs (see below)
**Describe alternatives you've considered**
- Do nothing.
- Use another UI.
**Additional context**
This is how it's implemented in https://chat.deepseek.com
Collapsed:
Expanded:
<!-- gh-comment-id:2608173414 -->
@silentoplayz commented on GitHub (Jan 22, 2025):
Addressed in the latest version released - https://github.com/open-webui/open-webui/releases/tag/v0.5.5
Wow, that was the fastest turnaround time for any issue I ever reported 😅 Yay to OpenWebUI! 🎉
<!-- gh-comment-id:2608175312 -->
@0x7CFE commented on GitHub (Jan 22, 2025):
Wow, that was the fastest turnaround time for any issue I ever reported 😅 Yay to OpenWebUI! 🎉
@silentoplayz Doesn't work with deepseek api. The reasoning content is separated in the response. See reasoning_content api-docs.deepseek.com
<!-- gh-comment-id:2609368203 -->
@konczdev commented on GitHub (Jan 23, 2025):
@silentoplayz Doesn't work with deepseek api. The reasoning content is separated in the response. See reasoning_content [api-docs.deepseek.com](https://api-docs.deepseek.com/guides/reasoning_model)
<!-- gh-comment-id:2610265649 -->
@silentoplayz commented on GitHub (Jan 23, 2025):
> [@silentoplayz](https://github.com/silentoplayz) Doesn't work with deepseek api. The reasoning content is separated in the response. See reasoning_content [api-docs.deepseek.com](https://api-docs.deepseek.com/guides/reasoning_model)
See the latest Open WebUI changelog: https://github.com/open-webui/open-webui/releases/tag/v0.5.6
@silentoplayz You think I didn't test before commenting on a closed issue? I explained the root cause too. Yes, it works with distilled models since the tag is part of the content in the response. However, the official DeepSeek API is different. Example from their api docs:
fromopenaiimportOpenAIclient=OpenAI(api_key="<DeepSeek API Key>",base_url="https://api.deepseek.com")# Round 1messages=[{"role":"user","content":"9.11 and 9.8, which is greater?"}]response=client.chat.completions.create(model="deepseek-reasoner",messages=messages,stream=True)reasoning_content=""content=""forchunkinresponse:ifchunk.choices[0].delta.reasoning_content:reasoning_content+=chunk.choices[0].delta.reasoning_contentelse:content+=chunk.choices[0].delta.content# Round 2messages.append({"role":"assistant","content":content})messages.append({'role':'user','content':"How many Rs are there in the word 'strawberry'?"})response=client.chat.completions.create(model="deepseek-reasoner",messages=messages,stream=True)
<!-- gh-comment-id:2610432311 -->
@konczdev commented on GitHub (Jan 23, 2025):
@silentoplayz You think I didn't test before commenting on a closed issue? I explained the root cause too. Yes, it works with distilled models since the <think> tag is part of the content in the response. However, the official DeepSeek API is different. Example from their api docs:
```python
from openai import OpenAI
client = OpenAI(api_key="<DeepSeek API Key>", base_url="https://api.deepseek.com")
# Round 1
messages = [{"role": "user", "content": "9.11 and 9.8, which is greater?"}]
response = client.chat.completions.create(
model="deepseek-reasoner",
messages=messages,
stream=True
)
reasoning_content = ""
content = ""
for chunk in response:
if chunk.choices[0].delta.reasoning_content:
reasoning_content += chunk.choices[0].delta.reasoning_content
else:
content += chunk.choices[0].delta.content
# Round 2
messages.append({"role": "assistant", "content": content})
messages.append({'role': 'user', 'content': "How many Rs are there in the word 'strawberry'?"})
response = client.chat.completions.create(
model="deepseek-reasoner",
messages=messages,
stream=True
)
```
Confirmed. Works on distilled deepseek models but not with the official 'deepseek-reasoner' api model.
<!-- gh-comment-id:2610891892 -->
@bgeneto commented on GitHub (Jan 23, 2025):
Confirmed. Works on distilled deepseek models but not with the official 'deepseek-reasoner' api model.
I can also confirm that the deepseek reasoner api model isn't displaying the "thought" process but the local distilled models are.
<!-- gh-comment-id:2614232610 -->
@TheLocalLab commented on GitHub (Jan 26, 2025):
I can also confirm that the deepseek reasoner api model isn't displaying the "thought" process but the local distilled models are.
However it can be achieved by Pipe functions. This one imitates <think></think> block from a local model and does the job for me: https://openwebui.com/f/zgccrui/deepseek_r1
<!-- gh-comment-id:2614488444 -->
@Yundin commented on GitHub (Jan 26, 2025):
Seems like there won't be an official support of this feature in the current state of DeepSeek API because its json differs from OpenAI API specs: https://github.com/open-webui/open-webui/issues/8749#issuecomment-2608126791
However it can be achieved by Pipe functions. This one imitates `<think></think>` block from a local model and does the job for me: https://openwebui.com/f/zgccrui/deepseek_r1
Would be good to communicate this caveat more prominently so we don't get confused thinking things are broken or we need to fix something 💙
<!-- gh-comment-id:2614737902 -->
@garymoon commented on GitHub (Jan 27, 2025):
Would be good to communicate this caveat more prominently so we don't get confused thinking things are broken or we need to fix something 💙
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @0x7CFE on GitHub (Jan 22, 2025).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/8769
Feature Request
Is your feature request related to a problem? Please describe.
Since reasoning models are now mainstream, we are now often facing an issue of too many reasoning tokens thrown right in our face. This can flood the UI and make your laptop levitating because of coolers blasting at 100% due to UI update cycle. But the issue here is mostly about UI clutter than performance.
Describe the solution you'd like
I'd like to have a foldable reasoning tokens widget in OpenWebUI, much like in other UIs (see below)
Describe alternatives you've considered
Additional context
This is how it's implemented in https://chat.deepseek.com
Collapsed:
Expanded:
@silentoplayz commented on GitHub (Jan 22, 2025):
Addressed in the latest version released - https://github.com/open-webui/open-webui/releases/tag/v0.5.5
@0x7CFE commented on GitHub (Jan 22, 2025):
Wow, that was the fastest turnaround time for any issue I ever reported 😅 Yay to OpenWebUI! 🎉
@konczdev commented on GitHub (Jan 23, 2025):
@silentoplayz Doesn't work with deepseek api. The reasoning content is separated in the response. See reasoning_content api-docs.deepseek.com
@silentoplayz commented on GitHub (Jan 23, 2025):
See the latest Open WebUI changelog: https://github.com/open-webui/open-webui/releases/tag/v0.5.6
@konczdev commented on GitHub (Jan 23, 2025):
@silentoplayz You think I didn't test before commenting on a closed issue? I explained the root cause too. Yes, it works with distilled models since the tag is part of the content in the response. However, the official DeepSeek API is different. Example from their api docs:
@bgeneto commented on GitHub (Jan 23, 2025):
Confirmed. Works on distilled deepseek models but not with the official 'deepseek-reasoner' api model.
@TheLocalLab commented on GitHub (Jan 26, 2025):
I can also confirm that the deepseek reasoner api model isn't displaying the "thought" process but the local distilled models are.
@Yundin commented on GitHub (Jan 26, 2025):
Seems like there won't be an official support of this feature in the current state of DeepSeek API because its json differs from OpenAI API specs: https://github.com/open-webui/open-webui/issues/8749#issuecomment-2608126791
However it can be achieved by Pipe functions. This one imitates
<think></think>block from a local model and does the job for me: https://openwebui.com/f/zgccrui/deepseek_r1@garymoon commented on GitHub (Jan 27, 2025):
Would be good to communicate this caveat more prominently so we don't get confused thinking things are broken or we need to fix something 💙