Any use case of `dspy.ReAct` ?

[pavaris-pm]

i've just made a quick walkthrough with ReAct module in dspy, however, it seems like it is not well-documented, could anyone provide me an example of creating a react flow of this?

i've already search in this repo issues, what i've found is #276, where i do as follows:

import dspy

class GiveTime:
    name = "GiveTime"
    input_variable = "empty"
    desc = "takes an empty string and returns the current local time"

    def __init__(self, k=3):
        pass
   
    def __call__(self, *args, **kwargs):
        return datetime.datetime.now().strftime("%H:%M:%S")

then, to utilize dspy,

mytool = GiveTime()
gen = dspy.ReAct('question -> answer', tools=[mytool])

After that, i want to make it to generate an answer, however, what i've got is something with the keyword arguments

# Call the ReAct module on a particular input
question = 'What is the current time today?' 

kw = {
     # where should i referenced this
}

result = gen(question=question, **kw)

print(f"Question: {question}")
print(f"Final Predicted Answer (after ReAct process): {result.answer}")

Here is my current error, i leave my keyword argument to be blank since i'm not sure where we need to fill it from

AttributeError                            Traceback (most recent call last)
Cell In[76], line 6
      2 question = 'What is the color of the sky?' # this is a positional args
      3 kw={
      4     
      5 }
----> 6 result = gen(question=question, **kw)
      8 print(f"Question: {question}")
      9 print(f"Final Predicted Answer (after ReAct process): {result.answer}")

File /usr/local/lib/python3.10/dist-packages/dspy/primitives/program.py:29, in Module.__call__(self, *args, **kwargs)
     28 def __call__(self, *args, **kwargs):
---> 29     return self.forward(*args, **kwargs)

File /usr/local/lib/python3.10/dist-packages/dspy/predict/react.py:78, in ReAct.forward(self, **kwargs)
     74 args = {key: kwargs[key] for key in self.input_fields.keys() if key in kwargs}
     76 for hop in range(self.max_iters):
     77     # with dspy.settings.context(show_guidelines=(i <= 2)):
---> 78     output = self.react[hop](**args)
     80     if action_val := self.act(output, hop): break
     81     args.update(output)

File /usr/local/lib/python3.10/dist-packages/dspy/predict/predict.py:60, in Predict.__call__(self, **kwargs)
     59 def __call__(self, **kwargs):
---> 60     return self.forward(**kwargs)

File /usr/local/lib/python3.10/dist-packages/dspy/predict/predict.py:74, in Predict.forward(self, **kwargs)
     72 # If temperature is 0.0 but its n > 1, set temperature to 0.7.
     73 temperature = config.get("temperature", None)
---> 74 temperature = lm.kwargs['temperature'] if temperature is None else temperature
     76 num_generations = config.get("n", None)
     77 if num_generations is None:

AttributeError: 'NoneType' object has no attribute 'kwargs'

Thanks for all of your help in advance.

[neoxelox]

The error is telling you that you didn't set a default LM for DSPy. After importing DSPy, do:

lm = dspy.OpenAI(model='gpt-3.5-turbo-instruct', max_tokens=300)
dspy.configure(lm=lm)

Obviously you will need an OpenAI API key. You should set it in the environment variable OPENAI_API_KEY before running the notebook/script.

Additionally, as said in https://github.com/stanfordnlp/dspy/issues/276 , there is currently a bug in the ReAct module, and you will need to remove .passages from the line 67 in the file dspy/predict/react.py of your own local copy of the DSPy library.

BTW: *args and **kwargs are parameters that collect all the positional (* (list)) and keyword (** (dict)) arguments of a function, so you don't need to pass a **kw to your gen.