Fix asyncio in AsyncLLM to use the running event loop if any

[alvarobartt]

Description

This PR adds a hot-fix for the AsyncLLM class and subclasses, since the event_loop is always created, but in Colab there's already an active running event loop, so that should be reused, otherwise the AsyncLLM subclasses or any other asyncio code creating event_loops will fail in Colab with the following exception RuntimeError: Cannot run the event loop while another loop is running.

On top of that, an utils function named in_notebook has been included so as to identify whether the code is running in a notebook, and if so, use nest-asyncio so as to be able to successfully use asyncio.

Thanks @burtenshaw for reporting and @gabrielmbmb for the hints!

[gabrielmbmb]

LGTM!

[burtenshaw]

How did you test this? I merged it into the DEITA branch docs/deita-tutorial and tried it in the notebook but I got the same error.

I would expect this to work:

from distilabel.llm.openai import OpenAILLM
from distilabel.pipeline.local import Pipeline
from distilabel.steps.task.evol_instruct.base import EvolInstruct

pipeline = Pipeline(name="DEITA")

evol_instruction_complexity = EvolInstruct(
    name="evol_instruction_complexity",
    llm=OpenAILLM(model="gpt-3.5-turbo"),
    num_evolutions=5,
    store_evolutions=True,
    generate_answers=True,
    include_original_instruction=True,
    pipeline=pipeline,
)

next(evol_instruction_complexity.process(([{"instruction": "How many fish are there in a dozen fish?"}])))

[gabrielmbmb]

For me it works, but I needed to use nest_asyncio:

import nest_asyncio
from distilabel.llm.openai import OpenAILLM
from distilabel.pipeline.local import Pipeline
from distilabel.steps.task.evol_instruct.base import EvolInstruct

nest_asyncio.apply()

pipeline = Pipeline(name="DEITA")

evol_instruction_complexity = EvolInstruct(
    name="evol_instruction_complexity",
    llm=OpenAILLM(model="gpt-3.5-turbo"),
    num_evolutions=5,
    store_evolutions=True,
    generate_answers=True,
    include_original_instruction=True,
    pipeline=pipeline,
)

next(evol_instruction_complexity.process(([{"instruction": "How many fish are there in a dozen fish?"}])))

maybe we can add nest_asyncio as dependency and in distilabel/__init__.py check if running on ipython environment, and if so nest_asyncio.apply()