mirror/LocalAI
1
0
Fork 0
mirror of https://github.com/mudler/LocalAI.git synced 2026-05-06 10:40:08 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: add reasoning effort and metadata to template (#5981)

Browse Source 
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto
2025-08-06 21:56:05 +02:00
committed by GitHub
parent 6a8149e1fd
commit b9a25b16e6
6 changed files with 26 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
core/http/endpoints/openai/chat.go
+1 -1
View File
@@ -305,7 +305,7 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
// If we are using the tokenizer template, we don't need to process the messages
// unless we are processing functions
if !config.TemplateConfig.UseTokenizerTemplate || shouldUseFn {
predInput = evaluator.TemplateMessages(input.Messages, config, funcs, shouldUseFn)
predInput = evaluator.TemplateMessages(*input, input.Messages, config, funcs, shouldUseFn)
log.Debug().Msgf("Prompt (after templating): %s", predInput)
if config.Grammar != "" {
core/http/endpoints/openai/completion.go
+8 -4
View File
@@ -109,8 +109,10 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
predInput := config.PromptStrings[0]
templatedInput, err := evaluator.EvaluateTemplateForPrompt(templates.CompletionPromptTemplate, *config, templates.PromptTemplateData{
Input: predInput,
SystemPrompt: config.SystemPrompt,
Input: predInput,
SystemPrompt: config.SystemPrompt,
ReasoningEffort: input.ReasoningEffort,
Metadata: input.Metadata,
})
if err == nil {
predInput = templatedInput
@@ -160,8 +162,10 @@ func CompletionEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, e
for k, i := range config.PromptStrings {
templatedInput, err := evaluator.EvaluateTemplateForPrompt(templates.CompletionPromptTemplate, *config, templates.PromptTemplateData{
SystemPrompt: config.SystemPrompt,
Input: i,
SystemPrompt: config.SystemPrompt,
Input: i,
ReasoningEffort: input.ReasoningEffort,
Metadata: input.Metadata,
})
if err == nil {
i = templatedInput
core/http/endpoints/openai/edit.go
+5 -3
View File
@@ -47,9 +47,11 @@ func EditEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, evaluat
for _, i := range config.InputStrings {
templatedInput, err := evaluator.EvaluateTemplateForPrompt(templates.EditPromptTemplate, *config, templates.PromptTemplateData{
Input: i,
Instruction: input.Instruction,
SystemPrompt: config.SystemPrompt,
Input: i,
Instruction: input.Instruction,
SystemPrompt: config.SystemPrompt,
ReasoningEffort: input.ReasoningEffort,
Metadata: input.Metadata,
})
if err == nil {
i = templatedInput
core/schema/openai.go
+4
View File
@@ -183,6 +183,10 @@ type OpenAIRequest struct {
Backend string `json:"backend" yaml:"backend"`
ModelBaseName string `json:"model_base_name" yaml:"model_base_name"`
ReasoningEffort string `json:"reasoning_effort" yaml:"reasoning_effort"`
Metadata map[string]string `json:"metadata" yaml:"metadata"`
}
type ModelsDataResponse struct {
core/templates/evaluator.go
+5 -1
View File
@@ -21,6 +21,8 @@ type PromptTemplateData struct {
Instruction string
Functions []functions.Function
MessageIndex int
ReasoningEffort string
Metadata map[string]string
}
type ChatMessageTemplateData struct {
@@ -133,7 +135,7 @@ func (e *Evaluator) evaluateJinjaTemplateForPrompt(templateType TemplateType, te
return e.cache.evaluateJinjaTemplate(templateType, templateName, conversation)
}
func (e *Evaluator) TemplateMessages(messages []schema.Message, config *config.BackendConfig, funcs []functions.Function, shouldUseFn bool) string {
func (e *Evaluator) TemplateMessages(input schema.OpenAIRequest, messages []schema.Message, config *config.BackendConfig, funcs []functions.Function, shouldUseFn bool) string {
if config.TemplateConfig.JinjaTemplate {
var messageData []ChatMessageTemplateData
@@ -283,6 +285,8 @@ func (e *Evaluator) TemplateMessages(messages []schema.Message, config *config.B
SuppressSystemPrompt: suppressConfigSystemPrompt,
Input: predInput,
Functions: funcs,
ReasoningEffort: input.ReasoningEffort,
Metadata: input.Metadata,
})
if err == nil {
predInput = templatedInput
core/templates/evaluator_test.go
+3 -3
View File
@@ -219,7 +219,7 @@ var _ = Describe("Templates", func() {
for key := range chatMLTestMatch {
foo := chatMLTestMatch[key]
It("renders correctly `"+key+"`", func() {
templated := evaluator.TemplateMessages(foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
templated := evaluator.TemplateMessages(schema.OpenAIRequest{}, foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
Expect(templated).To(Equal(foo["expected"]), templated)
})
}
@@ -232,7 +232,7 @@ var _ = Describe("Templates", func() {
for key := range llama3TestMatch {
foo := llama3TestMatch[key]
It("renders correctly `"+key+"`", func() {
templated := evaluator.TemplateMessages(foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
templated := evaluator.TemplateMessages(schema.OpenAIRequest{}, foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
Expect(templated).To(Equal(foo["expected"]), templated)
})
}
@@ -245,7 +245,7 @@ var _ = Describe("Templates", func() {
for key := range jinjaTest {
foo := jinjaTest[key]
It("renders correctly `"+key+"`", func() {
templated := evaluator.TemplateMessages(foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
templated := evaluator.TemplateMessages(schema.OpenAIRequest{}, foo["messages"].([]schema.Message), foo["config"].(*config.BackendConfig), foo["functions"].([]functions.Function), foo["shouldUseFn"].(bool))
Expect(templated).To(Equal(foo["expected"]), templated)
})
}