issue with conversational agent chain: chain with more then one expected input

[devalexandre]

I'm writing some use cases, to document the use of langchaingo, but even looking at the examples some things are not clear, the use of template is a case, I tested both with llama2 and openai, and the result was the same error .

2024/03/11 23:02:43 run not supported in chain with more then one expected input
exit status 1

code:

package main

import (
    "context"
    "log"

    "github.com/tmc/langchaingo/agents"
    "github.com/tmc/langchaingo/chains"
    "github.com/tmc/langchaingo/llms/ollama"
    "github.com/tmc/langchaingo/prompts"
)

func main() {
    llm, err := ollama.New(ollama.WithModel("llama2"))
    if err != nil {
        log.Fatal(err)
    }

    promptInput := prompts.PromptTemplate{
        Template: "{{.Name}} are 36 years old and live in {{.City}}.",
        PartialVariables: map[string]any{
            "Name": "Bob",
            "City": "New York",
        },
        TemplateFormat: prompts.TemplateFormatGoTemplate,
    }

    agent, err := agents.Initialize(
        llm,
        nil,
        agents.ConversationalReactDescription,
        agents.WithPrompt(promptInput),
    )
    if err != nil {
        log.Fatal(err)
    }

    options := []chains.ChainCallOption{
        chains.WithTemperature(0.8),
    }

    if err != nil {
        log.Fatal(err)
    }

    res, err := chains.Run(context.Background(), agent, "What is the name of the person?", options...)
    if err != nil {
        log.Fatal(err)
    }

    log.Println(res)
}

[tmc]

Thanks for the report, I'm going to adjust the issue title and point out issues in the conversational agent chain.

[tmc]

Part of what's happening here is by supplying agents.WithPrompt ends up overwriting the prompt assembly logic in getConversationalPrompt so the contents sent to the model don't include any of the embedded templates.

[devalexandre]

any solution for it?