agents: internationalize and add Chinese prompt

agents/conversational.go

@@ -2,23 +2,19 @@ package agents
 
 import (
 	"context"
-	_ "embed"
 	"fmt"
 	"regexp"
 	"strings"
 
 	"github.com/tmc/langchaingo/callbacks"
 	"github.com/tmc/langchaingo/chains"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/llms"
 	"github.com/tmc/langchaingo/prompts"
 	"github.com/tmc/langchaingo/schema"
 	"github.com/tmc/langchaingo/tools"
 )
 
-const (
-	_conversationalFinalAnswerAction = "AI:"
-)
-
 // ConversationalAgent is a struct that represents an agent responsible for deciding
 // what to do or give the final output if the task is finished given a set of inputs
 // and previous steps taken.
@@ -34,6 +30,10 @@ type ConversationalAgent struct {
 	Tools []tools.Tool
 	// Output key is the key where the final output is placed.
 	OutputKey string
+	// FinalAnswer is the final answer in various languages.
+	FinalAnswer string
+	// Lang is the language the prompt will use.
+	Lang i18n.Lang
 	// CallbacksHandler is the handler for callbacks.
 	CallbacksHandler callbacks.Handler
 }
@@ -45,6 +45,7 @@ func NewConversationalAgent(llm llms.Model, tools []tools.Tool, opts ...Option)
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadConversationalTranslatable()
 
 	return &ConversationalAgent{
 		Chain: chains.NewLLMChain(
@@ -54,6 +55,8 @@ func NewConversationalAgent(llm llms.Model, tools []tools.Tool, opts ...Option)
 		),
 		Tools:            tools,
 		OutputKey:        options.outputKey,
+		FinalAnswer:      i18n.AgentsMustPhrase(options.lang, "conversational final answer"),
+		Lang:             options.lang,
 		CallbacksHandler: options.callbacksHandler,
 	}
 }
@@ -69,7 +72,7 @@ func (a *ConversationalAgent) Plan(
 		fullInputs[key] = value
 	}
 
-	fullInputs["agent_scratchpad"] = constructScratchPad(intermediateSteps)
+	fullInputs["agent_scratchpad"] = constructScratchPad(intermediateSteps, a.Lang)
 
 	var stream func(ctx context.Context, chunk []byte) error
 
@@ -84,7 +87,10 @@ func (a *ConversationalAgent) Plan(
 		ctx,
 		a.Chain,
 		fullInputs,
-		chains.WithStopWords([]string{"\nObservation:", "\n\tObservation:"}),
+		chains.WithStopWords([]string{
+			fmt.Sprintf("\n%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+			fmt.Sprintf("\n\t%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+		}),
 		chains.WithStreamingFunc(stream),
 	)
 	if err != nil {
@@ -117,22 +123,22 @@ func (a *ConversationalAgent) GetTools() []tools.Tool {
 	return a.Tools
 }
 
-func constructScratchPad(steps []schema.AgentStep) string {
+func constructScratchPad(steps []schema.AgentStep, lang i18n.Lang) string {
 	var scratchPad string
 	if len(steps) > 0 {
 		for _, step := range steps {
 			scratchPad += step.Action.Log
-			scratchPad += "\nObservation: " + step.Observation
+			scratchPad += fmt.Sprintf("\n%s %s", i18n.AgentsMustPhrase(lang, "observation"), step.Observation)
 		}
-		scratchPad += "\n" + "Thought:"
+		scratchPad += fmt.Sprintf("\n%s", i18n.AgentsMustPhrase(lang, "thought"))
 	}
 
 	return scratchPad
 }
 
 func (a *ConversationalAgent) parseOutput(output string) ([]schema.AgentAction, *schema.AgentFinish, error) {
-	if strings.Contains(output, _conversationalFinalAnswerAction) {
-		splits := strings.Split(output, _conversationalFinalAnswerAction)
+	if strings.Contains(output, a.FinalAnswer) {
+		splits := strings.Split(output, a.FinalAnswer)
 
 		finishAction := &schema.AgentFinish{
 			ReturnValues: map[string]any{
@@ -144,7 +150,9 @@ func (a *ConversationalAgent) parseOutput(output string) ([]schema.AgentAction,
 		return nil, finishAction, nil
 	}
 
-	r := regexp.MustCompile(`Action: (.*?)[\n]*Action Input: (.*)`)
+	action, actionInput := i18n.AgentsMustPhrase(a.Lang, "action"),
+		i18n.AgentsMustPhrase(a.Lang, "action input")
+	r := regexp.MustCompile(fmt.Sprintf(`%s (.*?)[\n]*%s (.*)`, action, actionInput))
 	matches := r.FindStringSubmatch(output)
 	if len(matches) == 0 {
 		return nil, nil, fmt.Errorf("%w: %s", ErrUnableToParseOutput, output)
@@ -155,15 +163,6 @@ func (a *ConversationalAgent) parseOutput(output string) ([]schema.AgentAction,
 	}, nil, nil
 }
 
-//go:embed prompts/conversational_prefix.txt
-var _defaultConversationalPrefix string //nolint:gochecknoglobals
-
-//go:embed prompts/conversational_format_instructions.txt
-var _defaultConversationalFormatInstructions string //nolint:gochecknoglobals
-
-//go:embed prompts/conversational_suffix.txt
-var _defaultConversationalSuffix string //nolint:gochecknoglobals
-
 func createConversationalPrompt(tools []tools.Tool, prefix, instructions, suffix string) prompts.PromptTemplate {
 	template := strings.Join([]string{prefix, instructions, suffix}, "\n\n")
 

agents/executor.go

@@ -36,6 +36,7 @@ func NewExecutor(agent Agent, opts ...Option) *Executor {
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadExecutorTranslatable()
 
 	return &Executor{
 		Agent:                   agent,

agents/mrkl.go

@@ -9,16 +9,12 @@ import (
 
 	"github.com/tmc/langchaingo/callbacks"
 	"github.com/tmc/langchaingo/chains"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/llms"
 	"github.com/tmc/langchaingo/schema"
 	"github.com/tmc/langchaingo/tools"
 )
 
-const (
-	_finalAnswerAction = "Final Answer:"
-	_defaultOutputKey  = "output"
-)
-
 // OneShotZeroAgent is a struct that represents an agent responsible for deciding
 // what to do or give the final output if the task is finished given a set of inputs
 // and previous steps taken.
@@ -32,6 +28,10 @@ type OneShotZeroAgent struct {
 	Tools []tools.Tool
 	// Output key is the key where the final output is placed.
 	OutputKey string
+	// FinalAnswer is the final answer in various languages.
+	FinalAnswer string
+	// Lang is the language the prompt will use.
+	Lang i18n.Lang
 	// CallbacksHandler is the handler for callbacks.
 	CallbacksHandler callbacks.Handler
 }
@@ -46,6 +46,7 @@ func NewOneShotAgent(llm llms.Model, tools []tools.Tool, opts ...Option) *OneSho
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadMrklTranslatable()
 
 	return &OneShotZeroAgent{
 		Chain: chains.NewLLMChain(
@@ -55,6 +56,8 @@ func NewOneShotAgent(llm llms.Model, tools []tools.Tool, opts ...Option) *OneSho
 		),
 		Tools:            tools,
 		OutputKey:        options.outputKey,
+		FinalAnswer:      i18n.AgentsMustPhrase(options.lang, "mrkl final answer"),
+		Lang:             options.lang,
 		CallbacksHandler: options.callbacksHandler,
 	}
 }
@@ -70,8 +73,8 @@ func (a *OneShotZeroAgent) Plan(
 		fullInputs[key] = value
 	}
 
-	fullInputs["agent_scratchpad"] = constructMrklScratchPad(intermediateSteps)
-	fullInputs["today"] = time.Now().Format("January 02, 2006")
+	fullInputs["agent_scratchpad"] = constructMrklScratchPad(intermediateSteps, a.Lang)
+	fullInputs["today"] = time.Now().Format(i18n.AgentsMustPhrase(a.Lang, "today format"))
 
 	var stream func(ctx context.Context, chunk []byte) error
 
@@ -86,7 +89,10 @@ func (a *OneShotZeroAgent) Plan(
 		ctx,
 		a.Chain,
 		fullInputs,
-		chains.WithStopWords([]string{"\nObservation:", "\n\tObservation:"}),
+		chains.WithStopWords([]string{
+			fmt.Sprintf("\n%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+			fmt.Sprintf("\n\t%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+		}),
 		chains.WithStreamingFunc(stream),
 	)
 	if err != nil {
@@ -119,21 +125,21 @@ func (a *OneShotZeroAgent) GetTools() []tools.Tool {
 	return a.Tools
 }
 
-func constructMrklScratchPad(steps []schema.AgentStep) string {
+func constructMrklScratchPad(steps []schema.AgentStep, lang i18n.Lang) string {
 	var scratchPad string
 	if len(steps) > 0 {
 		for _, step := range steps {
 			scratchPad += "\n" + step.Action.Log
-			scratchPad += "\nObservation: " + step.Observation + "\n"
+			scratchPad += fmt.Sprintf("\n%s %s\n", i18n.AgentsMustPhrase(lang, "observation"), step.Observation)
 		}
 	}
 
 	return scratchPad
 }
 
 func (a *OneShotZeroAgent) parseOutput(output string) ([]schema.AgentAction, *schema.AgentFinish, error) {
-	if strings.Contains(output, _finalAnswerAction) {
-		splits := strings.Split(output, _finalAnswerAction)
+	if strings.Contains(output, a.FinalAnswer) {
+		splits := strings.Split(output, a.FinalAnswer)
 
 		return nil, &schema.AgentFinish{
 			ReturnValues: map[string]any{
@@ -143,7 +149,15 @@ func (a *OneShotZeroAgent) parseOutput(output string) ([]schema.AgentAction, *sc
 		}, nil
 	}
 
-	r := regexp.MustCompile(`Action:\s*(.+)\s*Action Input:\s(?s)*(.+)`)
+	action, actionInput, observation := i18n.AgentsMustPhrase(a.Lang, "action"),
+		i18n.AgentsMustPhrase(a.Lang, "action input"),
+		i18n.AgentsMustPhrase(a.Lang, "observation")
+	var r *regexp.Regexp
+	if strings.Contains(output, observation) {
+		r = regexp.MustCompile(fmt.Sprintf(`%s\s*(.+)\s*%s\s(?s)*(.+)%s`, action, actionInput, observation))
+	} else {
+		r = regexp.MustCompile(fmt.Sprintf(`%s\s*(.+)\s*%s\s(?s)*(.+)`, action, actionInput))
+	}
 	matches := r.FindStringSubmatch(output)
 	if len(matches) == 0 {
 		return nil, nil, fmt.Errorf("%w: %s", ErrUnableToParseOutput, output)

agents/mrkl_prompt.go

@@ -8,29 +8,6 @@ import (
 	"github.com/tmc/langchaingo/tools"
 )
 
-const (
-	_defaultMrklPrefix = `Today is {{.today}}.
-Answer the following questions as best you can. You have access to the following tools:
-
-{{.tool_descriptions}}`
-
-	_defaultMrklFormatInstructions = `Use the following format:
-
-Question: the input question you must answer
-Thought: you should always think about what to do
-Action: the action to take, should be one of [ {{.tool_names}} ]
-Action Input: the input to the action
-Observation: the result of the action
-... (this Thought/Action/Action Input/Observation can repeat N times)
-Thought: I now know the final answer
-Final Answer: the final answer to the original input question`
-
-	_defaultMrklSuffix = `Begin!
-
-Question: {{.input}}
-{{.agent_scratchpad}}`
-)
-
 func createMRKLPrompt(tools []tools.Tool, prefix, instructions, suffix string) prompts.PromptTemplate {
 	template := strings.Join([]string{prefix, instructions, suffix}, "\n\n")
 

agents/markl_test.go → agents/mrkl_test.go

@@ -4,6 +4,7 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/require"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/schema"
 )
 
@@ -38,7 +39,11 @@ func TestMRKLOutputParser(t *testing.T) {
 		},
 	}
 
-	a := OneShotZeroAgent{}
+	lang := i18n.EN
+	a := OneShotZeroAgent{
+		FinalAnswer: i18n.AgentsMustPhrase(lang, "mrkl final answer"),
+		Lang:        lang,
+	}
 	for _, tc := range testCases {
 		actions, finish, err := a.parseOutput(tc.input)
 		require.ErrorIs(t, tc.expectedErr, err)

agents/openai_functions_agent.go

@@ -38,6 +38,7 @@ func NewOpenAIFunctionsAgent(llm llms.Model, tools []tools.Tool, opts ...Option)
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadOpenAIFunctionsTranslatable()
 
 	return &OpenAIFunctionsAgent{
 		LLM:              llm,