LangChain Expression Language (LCEL) RunnablePassthrough() 和 RunnablePassthrough.assign()

1. RunnablePassthrough()

https://python.langchain.com/docs/how_to/passthrough/

When composing chains with several steps, sometimes you will want to pass data from previous steps unchanged for use as input to a later step.

The RunnablePassthrough class allows you to do just this, and is typically is used in conjuction with a RunnableParallel to pass data through to a later step in your constructed chains.

通过 RunnablePassthrough() 传递数据时，允许保持输入不变，即透传数据。

from langchain_core.runnables import RunnableParallel, RunnablePassthrough

runnable = RunnableParallel(
    passed=RunnablePassthrough(),
    modified=lambda x: x["num"] + 1,
)

result = runnable.invoke({"num": 1})
print("result: ", result)  # {'passed': {'num': 1}, 'modified': 2}

/home/yongqiang/miniconda3/bin/python /home/yongqiang/langchain_work/yongqiang/example.py 
result:  {'passed': {'num': 1}, 'modified': 2}

Process finished with exit code 0

As seen above, passed key was called with RunnablePassthrough() and so it simply passed on {'num': 1}.
使用 RunnablePassthrough() 将 {'num': 1} 原封不动地传递给 passed key。

We also set a second key in the map with modified. This uses a lambda to set a single value adding 1 to the num, which resulted in modified key with the value of 2.

2. RunnablePassthrough()

https://python.langchain.com/api_reference/core/runnables/langchain_core.runnables.passthrough.RunnablePassthrough.html

class RunnablePassthrough(RunnableSerializable[Other, Other]):

Runnable to passthrough inputs unchanged or with additional keys.
通过 RunnablePassthrough() 传递数据允许保持输入不变或添加额外的键 (key)。

This Runnable behaves almost like the identity function, except that it can be configured to add additional keys to the output, if the input is a dict.

The examples below demonstrate this Runnable works using a few simple chains. The chains rely on simple lambdas to make the examples easy to execute and experiment with.

from langchain_core.runnables import (
    RunnableLambda,
    RunnableParallel,
    RunnablePassthrough,
)

runnable1 = RunnableParallel(
    origin=RunnablePassthrough(),
    modified=lambda x: x + 1
)

result1 = runnable1.invoke(1)
print("result1: ", result1)  # {'origin': 1, 'modified': 2}

runnable2 = RunnableParallel(
    original=RunnablePassthrough(),
    parsed=lambda text: text[::-1]
)

result2 = runnable2.invoke("yongqiang")
print("result2: ", result2)  # {'original': 'yongqiang', 'parsed': 'gnaiqgnoy'}

/home/yongqiang/miniconda3/bin/python /home/yongqiang/langchain_work/yongqiang/example.py 
result1:  {'origin': 1, 'modified': 2}
result2:  {'original': 'yongqiang', 'parsed': 'gnaiqgnoy'}

Process finished with exit code 0

• Include input dict in output dict

from langchain_core.runnables import (
    RunnableLambda,
    RunnableParallel,
    RunnablePassthrough,
)

runnable1 = RunnableLambda(lambda x: x["foo"] + 7)

chain = RunnableParallel(bar=runnable1, baz=RunnablePassthrough())

result = chain.invoke({"foo": 10})
print("result:", result)

/home/yongqiang/miniconda3/bin/python /home/yongqiang/langchain_work/yongqiang/example.py 
result: {'bar': 17, 'baz': {'foo': 10}}

Process finished with exit code 0

2.1. RunnablePassthrough.assign()

In some cases, it may be useful to pass the input through while adding some keys to the output. In this case, you can use the assign method.

from langchain_core.runnables import RunnableParallel, RunnablePassthrough

runnable = RunnableParallel(
    extra=RunnablePassthrough.assign(multi=lambda x: x["num"] * 3),
    modified=lambda x: x["num"] + 1,
)

result = runnable.invoke({"num": 1})
print("result: ", result)  # {'extra': {'num': 1, 'multi': 3}, 'modified': 2}

/home/yongqiang/miniconda3/bin/python /home/yongqiang/langchain_work/yongqiang/example.py 
result:  {'extra': {'num': 1, 'multi': 3}, 'modified': 2}

Process finished with exit code 0

• 输入是 {"num": 1}，输入被传递到 RunnableParallel()，RunnableParallel() 与输入一起并行调用可运行对象，RunnableParallel() 中的两个操作是并行执行的。
• RunnablePassthrough.assign() 会保留输入字典 {"num": 1}，并新增一个名为 multi 的 key，其 value 是 lambda x: x["num"] * 3，即 3。因此，结果是 {'num': 1, 'multi': 3}，{'num': 1, 'multi': 3} 作为键 extra 的 value。
• 同时，lambda x: x["num"] + 1 的结果是 2，因为 lambda 从输入中提取键 "num" 的 value，并加 1。

因此，最终的结果是 {'extra': {'num': 1, 'multi': 3}, 'modified': 2}。

• Merge input and output dicts

from langchain_core.runnables import RunnableLambda, RunnablePassthrough

runnable1 = RunnableLambda(lambda x: x["foo"] + 7)

chain = RunnablePassthrough.assign(bar=runnable1)

result = chain.invoke({"foo": 10})
print("result:", result)

/home/yongqiang/miniconda3/bin/python /home/yongqiang/langchain_work/yongqiang/example.py 
result: {'foo': 10, 'bar': 17}

Process finished with exit code 0

References

[1] Yongqiang Cheng, https://yongqiang.blog.csdn.net/