Looping pipelines with a PromptBuilder is not running

[bilgeyucel]

Describe the bug
If the PrompBuilder is the first component in looping pipelines, the pipeline gets stuck in a loop (might not be running at all) even when max_loops_allowed is set to a small number.

Error message
There's no error message. The pipeline is not outputting anything however the colab cell with pipeline.run doesn't stop running.

Expected behavior
Looping pipelines run without any problem.

Additional context
Relevant discussion: #7554

To Reproduce
Run tutorial 28: https://haystack.deepset.ai/tutorials/28_structured_output_with_loop

FAQ Check

• Have you had a look at our new FAQ page?

System:

• OS: google colab
• GPU/CPU: CPU
• Haystack version (commit or version number): Got the error in three versions: 2.0.0, 2.0.1, 2.1.0-rc
• DocumentStore: -
• Reader: -
• Retriever: -

[bilgeyucel]

Feel free to add more context @silvanocerza

[nateostro]

Encountered this yesterday. Fixed it this morning. Will link the PR here.

[silvanocerza]

Investigated a bit more and found out that this is actually already solved with #7531.
Tutorial is running fine on Haystack main.

Closing.

[Ovank]

@silvanocerza , I am still observing this issue, I did a fresh clone of main and re-ran the tutorial code with my local LLM , still process going in infinite loop . Code is also not generating any log . Can you help with the log generation issue ?

[bilgeyucel]

@Ovank Thanks for the update. I tried to run tutorial 28 on google colab just now and it worked without any problems.
Which Generator are you using in your pipeline?

[Ovank]

@bilgeyucel , I am using HuggingFaceLocalGenerator to create generator object , Sharing snippet from my code.

from haystack.components.generators.chat import HuggingFaceLocalChatGenerator 
from haystack.components.generators import HuggingFaceLocalGenerator
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

model_id = "./path_to_checkpoint/llama-2-7b-chat-hf/"
tokenizer = AutoTokenizer.from_pretrained(model_id,use_fast=False )

generator = HuggingFaceLocalGenerator(
    task = "text-generation", 
    huggingface_pipeline_kwargs = { 
        "model" : model_id, 
        "tokenizer" : tokenizer
    },
    generation_kwargs = {
            "temperature": 0.9,
            "max_new_tokens": 2096,
    }
)

llm = generator

[airibarne]

Any other news on this? Seems not solved in current Haystack version (2.7.0), running the tutorial as-is will result in RuntimeWarning: Pipeline is stuck running in a loop:

• Tutorial: https://haystack.deepset.ai/tutorials/28_structured_output_with_loop
• Self-contained script: https://gist.github.com/airibarne/452ef4901cb23b745c86c6238fe1a701