Fix: Add instance check before casting ChatGeneration object

[Youngrok123]

My case

• When receiving responses from headline and summary extractors, etc., they are delivered as a Generation object, not ChatGeneration.

{
  "generations": [
    [
      {
        "text": "{\"text\": \"ABCD\"}",
        "generation_info": null,
        "type": "Generation"
      }
    ]
  ],
  "llm_output": null,
  "run": null,
  "type": "LLMResult"
}

Occurrence message

• unable to apply transformation: 'Generation' object has no attribute 'message'

How to fix

• Add code to first check if it is an instance of that object before casting to a ChatGeneration object.

[jjmachan]

thanks again @Youngrok123 🙂 ❤️
btw which model are you using?

[Youngrok123]

thanks, @jjmachan
I am using 'neuralmagic/Meta-Llama-3.1-70B-Instruct-FP8' model on vLLM.
And as shown in the blog below, it inherits and implements the Langchain LLM class to communicate with the LLM.
LangChain: How to Set a Custom LLM Wrapper

If an exception occurs when generating a testset, the process will stop or normal results will not be produced, so handling the exception is necessary.

[bay-mini]

same problem

[jjmachan]

@bay-mini @Youngrok123 released with v0.2.4 🙂