Skip to content

feat(openresponses): Support reasoning blocks #8133

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mudler merged 4 commits into from
Jan 20, 2026
Merged

feat(openresponses): Support reasoning blocks #8133

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5f4b0da
feat(openresponses): support reasoning blocks
mudler Jan 20, 2026
8d8ab53
allow to disable reasoning, refactor common logic
mudler Jan 20, 2026
3620168
Add option to only strip reasoning
mudler Jan 20, 2026
3043572
Add configurations for custom reasoning tokens
mudler Jan 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 7 additions & 21 deletions core/http/endpoints/openai/chat.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,7 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
} else {
template = s
}
thinkingStartToken := reason.DetectThinkingStartToken(template)
thinkingStartToken := reason.DetectThinkingStartToken(template, &config.ReasoningConfig)

// Track accumulated content for reasoning extraction
accumulatedContent := ""
Expand All @@ -56,12 +56,8 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator

_, _, err := ComputeChoices(req, s, config, cl, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, tokenUsage backend.TokenUsage) bool {
accumulatedContent += s
content := accumulatedContent
// Prepend thinking token if needed, then extract reasoning
if config.ReasoningConfig.DisableReasoningTagPrefill == nil || !*config.ReasoningConfig.DisableReasoningTagPrefill {
content = reason.PrependThinkingTokenIfNeeded(content, thinkingStartToken)
}
currentReasoning, cleanedContent := reason.ExtractReasoning(content)

currentReasoning, cleanedContent := reason.ExtractReasoningWithConfig(accumulatedContent, thinkingStartToken, config.ReasoningConfig)

// Calculate new reasoning delta (what we haven't emitted yet)
var reasoningDelta *string
Expand Down Expand Up @@ -140,7 +136,7 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
} else {
template = prompt
}
thinkingStartToken := reason.DetectThinkingStartToken(template)
thinkingStartToken := reason.DetectThinkingStartToken(template, &config.ReasoningConfig)

result := ""
lastEmittedCount := 0
Expand Down Expand Up @@ -254,12 +250,7 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
return err
}
// Prepend thinking token if needed, then extract reasoning before processing tool calls
resultWithToken := result
if config.ReasoningConfig.DisableReasoningTagPrefill == nil || !*config.ReasoningConfig.DisableReasoningTagPrefill {
resultWithToken = reason.PrependThinkingTokenIfNeeded(result, thinkingStartToken)
}
reasoning, cleanedResult := reason.ExtractReasoning(resultWithToken)
result = cleanedResult
reasoning, result := reason.ExtractReasoningWithConfig(result, thinkingStartToken, config.ReasoningConfig)

textContentToReturn = functions.ParseTextContent(result, config.FunctionsConfig)
result = functions.CleanupLLMResult(result, config.FunctionsConfig)
Expand Down Expand Up @@ -652,18 +643,13 @@ func ChatEndpoint(cl *config.ModelConfigLoader, ml *model.ModelLoader, evaluator
} else {
template = predInput
}
thinkingStartToken := reason.DetectThinkingStartToken(template)
thinkingStartToken := reason.DetectThinkingStartToken(template, &config.ReasoningConfig)

xlog.Debug("Thinking start token", "thinkingStartToken", thinkingStartToken, "template", template)

tokenCallback := func(s string, c *[]schema.Choice) {
// Prepend thinking token if needed, then extract reasoning from the response
sWithToken := s
if config.ReasoningConfig.DisableReasoningTagPrefill == nil || !*config.ReasoningConfig.DisableReasoningTagPrefill {
sWithToken = reason.PrependThinkingTokenIfNeeded(s, thinkingStartToken)
}
reasoning, cleanedS := reason.ExtractReasoning(sWithToken)
s = cleanedS
reasoning, s := reason.ExtractReasoningWithConfig(s, thinkingStartToken, config.ReasoningConfig)

if !shouldUseFn {
// no function is called, just reply and use stop as finish reason
Expand Down
Loading
Loading