dosubot[bot]
commented
3 months ago Hey there, @mamlukishay! I'm here to assist you with your LangChain.js issue. Let's squash this bug together! 🐞
To handle schema validation errors in LangChain.js and ensure that the ToolInputParsingException is recognized so that the chain does not proceed incorrectly, you can use the call method of the StructuredTool class. This method parses the input according to the schema and throws a ToolInputParsingException if the input does not match the expected schema. Here is an example of how you can handle this:
import { z } from "zod";
import { ToolInputParsingException, StructuredTool } from "path/to/langchain-core/src/tools";
// Define your schema using Zod
const briefSchema = z.object({
requestedKPIs: z.nativeEnum(CampaignOptimizationKPI).optional(),
});
// Create a custom tool by extending StructuredTool
class SubmitCampaignBriefTool extends StructuredTool<typeof briefSchema> {
schema = briefSchema;
async _call(arg: z.output<typeof briefSchema>): Promise<string> {
// Your tool logic here
return `OK, brief updated`;
}
}
// Instantiate your tool
const tool = new SubmitCampaignBriefTool();
// Example usage
async function runTool(input: unknown) {
try {
const result = await tool.call(input);
console.log(result);
} catch (e) {
if (e instanceof ToolInputParsingException) {
console.error("Input parsing error:", e.message);
// Handle the error appropriately
} else {
console.error("Unexpected error:", e);
}
}
}
// Test with invalid input
runTool({ requestedKPIs: 'ctr' });Additionally, you can handle parsing errors more gracefully by using the handleParsingErrors option in the AgentExecutor class. Here is an example:
import { z } from "zod";
import { ChatOpenAI } from "@langchain/openai";
import { AgentExecutor, createOpenAIFunctionsAgent } from "langchain/agents";
import { pull } from "langchain/hub";
import { DynamicStructuredTool } from "@langchain/core/tools";
const model = new ChatOpenAI({ temperature: 0.1 });
const tools = [
new DynamicStructuredTool({
name: "submit-campaign-brief",
description: "submit incremental/full updates about the campaign brief object",
schema: briefSchema,
func: async (input) => {
console.log(input);
return 'OK, brief updated';
},
}),
];
// Get the prompt to use - you can modify this!
const prompt = await pull<ChatPromptTemplate>(
"hwchase17/openai-functions-agent"
);
const agent = await createOpenAIFunctionsAgent({
llm: model,
tools,
prompt,
});
const agentExecutor = new AgentExecutor({
agent,
tools,
verbose: true,
handleParsingErrors:
"Please try again, paying close attention to the allowed enum values",
});
console.log("Loaded agent.");
const input = `Set a reminder to renew our online property ads next week.`;
console.log(`Executing with input "${input}"...`);
const result = await agentExecutor.invoke({ input });
console.log({ result });In this example, the handleParsingErrors option provides a custom message to guide the user in correcting their input, ensuring the chain does not proceed incorrectly [1][2].
To continue talking to Dosu, mention @dosu.
mamlukishay
obround
Checked other resources
Example Code
The following tool always returns as successful (with empty observation), even if it fails schema validation.
Error Message and Stack Trace (if applicable)
No response
Description
I'm invoking a langchain tool through an agent executor, with a wrong schema. Langchain doesn't identify the error as a
ToolInputParsingException, therefore my chain proceeds and I get wrong results.This is my tool definition (simplified for the example):
The following input fails schema validation:
This is the stack trace:
This error is checked in executor.js for its instance type, but not recognized as the correct instance.
With this error being swallowed and never thrown, I don't have no ways to handle schema validation problems. This is bad, Please assist. 🙏
System Info
langchain@0.2.12 node v20.11.0 yarn v1.22.22