Using the model with Langchain agents sometimes breaks because the special end of sequence token "</s>" is added at the end of the output
I was playing the model and langchain agent. I noticed after some steps the model adds the special token "< /s>" at the end of the output and makes the agent breaks. The root. I am not an expert so that I did get the reason why the model adds it. Anyway, I added a simple parsing rule to remove this string from the output and now it works pretty well. I just share this if someone in the community was struggling with this model when combined with langchain agent.
Here is a sample example (I added a white space in the token otherwise it does not displays)
Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
Instruction:
Write a poem about a squirrel who like soccer
Response: < /s>
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
namespace _031_MinimumNumberOfOperationsToReachTargetValue
{
class Program
{
static void Main(string[] args)
{
int n = int.Parse(Console.ReadLine());
int target = int.Parse(Console.ReadLine()) - (n / 2);
Console.WriteLine("The minimum number of operations to reach " + target + " from " + n + " is " + GetMinOperations(target, n));
}
private static int GetMinOperations(int target, int n)
{
if (n == 0 || target < 0) return 0;
List<int> list
forget my dump remarks above. The issue was first my lack of knowledge on how it works and second a misconfiguration of the text generation webui API.
OK, glad you got it sorted