How to route to other type Chains, such as QA Chain, instead of LLMChain in destination_train

[flashslothless]

I am trying to router request to 2 Chains by using RouterChains:

1. LLM Chains
2. QA1 Chains

It works when all destiation_trains is LLMChain, but when I set one to be QA Chains, it always got errors below:

ValidationError Traceback (most recent call last)
Cell In[21], line 12
6 router_prompt = PromptTemplate(
7 template=router_template,
8 input_variables=["input"],
9 output_parser=RouterOutputParser(),
10 )
11 router_chain = LLMRouterChain.from_llm(llm, router_prompt)
---> 12 chain = MultiPromptChain(router_chain=router_chain, destination_chains=destination_chains, default_chain=default_chain, verbose=True)

File ~/anaconda3/envs/python3/lib/python3.10/site-packages/langchain/load/serializable.py:61, in Serializable.init(self, **kwargs)
60 def init(self, **kwargs: Any) -> None:
---> 61 super().init(**kwargs)
62 self.lc_kwargs = kwargs

File ~/anaconda3/envs/python3/lib/python3.10/site-packages/pydantic/main.py:341, in pydantic.main.BaseModel.init()
ValidationError: 6 validation errors for MultiPromptChain
destination_chains -> QA1 -> prompt
field required (type=value_error.missing)
destination_chains -> QA1 -> llm
field required (type=value_error.missing)
destination_chains -> QA1 -> combine_documents_chain
extra fields not permitted (type=value_error.extra)
destination_chains -> QA1 -> input_key
extra fields not permitted (type=value_error.extra)
destination_chains -> QA1 -> retriever
extra fields not permitted (type=value_error.extra)
destination_chains -> QA1 -> return_source_documents
extra fields not permitted (type=value_error.extra)

---

QA1 Chain can work well when I test it seperately, and all parameters are set.
Any advises are apprecited!

[flashslothless]

Cause:

MultiRouteChain is base abastract class, does not include implemtation.
but the sample MultiPromptChain set the desitnation to be LLMChain only.
destination_chains: Mapping[str, LLMChain]

Solution:

If you need to route to different type of destination chain instead of LLM, you need to extend MultiRouteChain to your class.
In my case, define a new class :

class MultitypeDestRouteChain(MultiRouteChain) : and set destination type be be Chain, will fix it.

"""Chain for deciding a destination chain and the input to it."""
    destination_chains: Mapping[str, Chain]

[mohit19906]

I am also getting the same error even though both my destination chains are the QA Chains. Below is the error -

pydantic.error_wrappers.ValidationError: 6 validation errors for MultiPromptChain destination_chains -> memberInstall -> prompt field required (type=value_error.missing) destination_chains -> memberInstall -> llm field required (type=value_error.missing) destination_chains -> memberInstall -> combine_documents_chain extra fields not permitted (type=value_error.extra) destination_chains -> memberInstall -> input_key extra fields not permitted (type=value_error.extra) destination_chains -> memberInstall -> retriever extra fields not permitted (type=value_error.extra) destination_chains -> memberInstall -> return_source_documents extra fields not permitted (type=value_error.extra)

Can you please elaborate your solution and provide the exact line of code. I am a new bee and does not understand what you said. It will be really helpful.

[flashslothless]

The key is you need to define your implementation of MultiRouteChain.
Below sample use MultiRouteChain to route three kinds of Chains: LLMChain, APIChain, RetrievalQA.

Implemetation of Multitype dest router:

class MultitypeDestRouteChain(MultiRouteChain) :
    """A multi-route chain that uses an LLM router chain to choose amongst prompts."""

    router_chain: RouterChain
    """Chain for deciding a destination chain and the input to it."""
    destination_chains: Mapping[str, Chain]
    """Map of name to candidate chains that inputs can be routed to."""
    default_chain: LLMChain
    """Default chain to use when router doesn't map input to one of the destinations."""

    @property
    def output_keys(self) -> List[str]:
        return ["text"]

Init three types Chains:

common_document_generation_chain = LLMChain(...)
koopage_document_qa_chain = RetrievalQA.from_chain_type(...)
movie_tv_chain = APIChain.from_llm_and_api_docs(...)

Init the MultitypeDestRouteChain

destination_chains = {}
destination_chains["common_document_generation"] = common_document_generation_chain
destination_chains["page_document_qa"] = page_document_qa_chain
destination_chains["movie_tv_search"] = movie_tv_chain

chain = MultitypeDestRouteChain(
    router_chain=router_chain,
    destination_chains=destination_chains,
    default_chain=default_chain,
    verbose=True,
)

chain.run(userPrompt)

[wolfassi123]

I am facing the same issue but using Agents. I have an agent with several routes, mainly route with LLM Chains and one with a ConversationalRelationChain. Have been struggling in running the agent properly as each chain takes a different output type. Tried LOTR (Lord of all Retrievers) but it is still not working. Anyone with some experience with such a problem?