Add support for reasoning_content field in chat completion messages for DeepSeek R1 #925

goodenough227 · 2025-01-23T09:38:30Z

feat: Add support for reasoning_content field in chat completion messages

This PR adds support for the reasoning_content field in the ChatCompletionMessage struct to accommodate models that return their reasoning process separately from the main content. This feature is particularly useful for:

DeepSeek R1 model

Both models return their step-by-step reasoning process through this field.

Changes:

Added reasoning_content field to ChatCompletionMessage struct
Updated JSON marshaling/unmarshaling to handle the new field
Field is optional (omitempty) to maintain backwards compatibility

Example response structure:

{
    "role": "assistant",
    "reasoning_content": "1. First, I analyze the question...\n2. Based on the given information...",
    "content": "The final answer is..."
}

codecov · 2025-01-23T11:59:44Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.88%. Comparing base (774fc9d) to head (319a8ea).
Report is 88 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #925      +/-   ##
==========================================
+ Coverage   98.46%   98.88%   +0.42%     
==========================================
  Files          24       27       +3     
  Lines        1364     1790     +426     
==========================================
+ Hits         1343     1770     +427     
+ Misses         15       14       -1     
  Partials        6        6

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

sashabaranov · 2025-01-23T17:39:30Z

@goodenough227 could you please point to the OpenAI API reference for this one? Or is this field only supported by DeepSeek currently?

yayanet · 2025-01-26T06:34:12Z

@goodenough227 could you please point to the OpenAI API reference for this one? Or is this field only supported by DeepSeek currently?

I also have a need for this feature.
The reasoning_content field is currently specific to deepseek-reasoner model (docs)

goodenough227 · 2025-01-26T09:51:03Z

@goodenough227 could you please point to the OpenAI API reference for this one? Or is this field only supported by DeepSeek currently?

sorry for this wrong description generated by AI code hepler. "reasoning_content" field is only supported by deepseek R1 currently.

sashabaranov · 2025-01-31T19:12:37Z

Originally, this repository was targeted on the official OpenAI API. But, in the context of the official client popping up — I'm willing to merge this reasoning_content field that OpenAI API might not support.

My only ask here is: can we please clearly indicate that this field is not supported by OpenAI? Ideally, it should be obvious from the variable name and from reading the docs. Thank you!

goodenough227 · 2025-02-05T03:09:38Z

Originally, this repository was targeted on the official OpenAI API. But, in the context of the official client popping up — I'm willing to merge this reasoning_content field that OpenAI API might not support.

My only ask here is: can we please clearly indicate that this field is not supported by OpenAI? Ideally, it should be obvious from the variable name and from reading the docs. Thank you!

Get it. I have written a comment on this field.

# Conflicts: # chat.go # openai_test.go

goodenough227 · 2025-02-07T13:11:37Z

@sashabaranov I have added comments in a similar way as in the previous code, please review

He-hongjun · 2025-02-10T15:21:50Z

Hope to merge and support this as soon as possible. Thank you for your contribution! @sashabaranov @goodenough227

wangriyu · 2025-03-06T09:21:05Z

please review&approved, thank you @sashabaranov @goodenough227

xiaozhi747 · 2025-03-10T06:22:28Z

mark

agclqq · 2025-03-12T03:33:51Z

I am exactly what I want, please approve, thank you @sashabaranov @goodenough227

kmesiab · 2025-03-12T03:40:22Z

openai_test.go

@@ -29,7 +29,7 @@ func setupAzureTestServer() (client *openai.Client, server *test.ServerTest, tea

 // numTokens Returns the number of GPT-3 encoded tokens in the given text.
 // This function approximates based on the rule of thumb stated by OpenAI:
-// https://beta.openai.com/tokenizer/
+// https://beta.openai.com/tokenizer.


Why is this changing?

This is to pass the golint check. When using golint for inspection, golint will require this to be changed to "."

kmesiab · 2025-03-12T03:41:21Z

chat_test.go

-				Content: "Hello!",
+				Role:             openai.ChatMessageRoleUser,
+				Content:          "Hello!",
+				ReasoningContent: "hi!",


Does this deserve its own test in isolation?

This is currently targeting a specific niche model.

I have removed the ReasoningContent section here but did not add the constants and other test functions for the deepseek model, as it seems that, from your perspective, go-openai is primarily intended for OpenAI models. Please share your thoughts on this so I can align it with my subsequent changes.

go-openai is primarily intended for OpenAI models.

This is my personal stance, yes. The addition of support for DeepSeek feels like a knee-jerk reaction. While I appreciate the flexibility your addition provides, it is still a divergence from the spirit of the repo. The o1 models, and the future path of OpenAI's API may very well introduce a competing field.

The compromise, if you will, is to limit the coupling, like here in the tests.

I understand. I have written test functions specifically for deepseek r1 using literal model names to minimize the invasiveness to the main codebase.

Additionally, I'd like to know if there are plans to support non-OpenAI official models in the future, as many new models (like deepseek open-source models) are using OpenAI's data structures. We are currently using our forked branch of go-openai in this way.

agclqq · 2025-03-12T04:59:49Z

I see that openai/openai-python has already supported ResponseReasoning, and I believe it will be supported soon in official client. Whether we use the current PR or build another one, I hope we can support it soon, thank you!

@sashabaranov @goodenough227 @goodenough227

goodenough227 · 2025-03-12T06:11:20Z

I see that openai/openai-python has already supported ResponseReasoning, and I believe it will be supported soon in official client. Whether we use the current PR or build another one, I hope we can support it soon, thank you!

@sashabaranov @goodenough227 @goodenough227
I believe it would be better to open a separate PR, as ResponseReasoning is a distinct data structure.

agclqq · 2025-03-13T03:42:42Z

Any other questions, rush

…ns, including support for reasoning content in responses.

goodenough227 added 3 commits January 21, 2025 13:11

support deepseek field "reasoning_content"

5297ba3

support deepseek field "reasoning_content"

1d922c6

Comment ends in a period (godot)

5fef03f

goodenough227 changed the title ~~Add support for reasoning_content field in chat completion messages for DeepSeek R1 and OpenAI O1~~ Add support for reasoning_content field in chat completion messages for DeepSeek R1 Jan 26, 2025

add comment on field reasoning_content

e025915

goodenough227 closed this Feb 5, 2025

goodenough227 force-pushed the master branch from e025915 to 45aa996 Compare February 5, 2025 12:47

Merge branch 'feat/reasoning-content'

1e313ed

# Conflicts: # chat.go # openai_test.go

goodenough227 reopened this Feb 5, 2025

goodenough227 and others added 4 commits February 5, 2025 21:00

fix go lint error

1105cfc

chore: trigger CI

09ab25f

Merge branch 'sashabaranov:master' into master

c31b505

make field "content" in MarshalJSON function omitempty

5273861

jchen42703 mentioned this pull request Feb 19, 2025

Have you considered supporting the reasoning_content field for the Deepseek R1 model in the ChatCompletionMessage and ChatCompletionStreamChoiceDelta structs? #935

Open

kmesiab reviewed Mar 12, 2025

View reviewed changes

goodenough227 and others added 2 commits March 12, 2025 13:29

Merge branch 'sashabaranov:master' into master

e222662

remove reasoning_content in TestO1ModelChatCompletions func

fd8334a

goodenough227 added 3 commits March 13, 2025 14:01

feat: Add test and handler for deepseek-reasoner chat model completio…

a450fb8

…ns, including support for reasoning content in responses.

feat: Add test and handler for deepseek-reasoner chat model completio…

6ef63e2

…ns, including support for reasoning content in responses.

feat: Add test and handler for deepseek-reasoner chat model completio…

319a8ea

…ns, including support for reasoning content in responses.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for reasoning_content field in chat completion messages for DeepSeek R1 #925

Add support for reasoning_content field in chat completion messages for DeepSeek R1 #925

goodenough227 commented Jan 23, 2025 •

edited

Loading

codecov bot commented Jan 23, 2025 •

edited

Loading

sashabaranov commented Jan 23, 2025

yayanet commented Jan 26, 2025

goodenough227 commented Jan 26, 2025 •

edited

Loading

sashabaranov commented Jan 31, 2025

goodenough227 commented Feb 5, 2025 •

edited

Loading

goodenough227 commented Feb 7, 2025

He-hongjun commented Feb 10, 2025

wangriyu commented Mar 6, 2025

xiaozhi747 commented Mar 10, 2025

agclqq commented Mar 12, 2025

kmesiab Mar 12, 2025

goodenough227 Mar 12, 2025

kmesiab Mar 12, 2025

goodenough227 Mar 12, 2025

kmesiab Mar 12, 2025 •

edited

Loading

goodenough227 Mar 13, 2025

agclqq commented Mar 12, 2025

goodenough227 commented Mar 12, 2025

agclqq commented Mar 13, 2025

Add support for reasoning_content field in chat completion messages for DeepSeek R1 #925

Are you sure you want to change the base?

Add support for reasoning_content field in chat completion messages for DeepSeek R1 #925

Conversation

goodenough227 commented Jan 23, 2025 • edited Loading

codecov bot commented Jan 23, 2025 • edited Loading

Codecov Report

sashabaranov commented Jan 23, 2025

yayanet commented Jan 26, 2025

goodenough227 commented Jan 26, 2025 • edited Loading

sashabaranov commented Jan 31, 2025

goodenough227 commented Feb 5, 2025 • edited Loading

goodenough227 commented Feb 7, 2025

He-hongjun commented Feb 10, 2025

wangriyu commented Mar 6, 2025

xiaozhi747 commented Mar 10, 2025

agclqq commented Mar 12, 2025

kmesiab Mar 12, 2025

Choose a reason for hiding this comment

goodenough227 Mar 12, 2025

Choose a reason for hiding this comment

kmesiab Mar 12, 2025

Choose a reason for hiding this comment

goodenough227 Mar 12, 2025

Choose a reason for hiding this comment

kmesiab Mar 12, 2025 • edited Loading

Choose a reason for hiding this comment

goodenough227 Mar 13, 2025

Choose a reason for hiding this comment

agclqq commented Mar 12, 2025

goodenough227 commented Mar 12, 2025

agclqq commented Mar 13, 2025

goodenough227 commented Jan 23, 2025 •

edited

Loading

codecov bot commented Jan 23, 2025 •

edited

Loading

goodenough227 commented Jan 26, 2025 •

edited

Loading

goodenough227 commented Feb 5, 2025 •

edited

Loading

kmesiab Mar 12, 2025 •

edited

Loading