[BUG] [RAG] Error when cloning with changed chunk_size

[richard-rfai]

Bug Description

When using the "clone-modify" IC Op to clone a RAG run with a changed chunk_size, whether through the IC Ops panel or directly with a call to the dispatcher, the cloned run will immediately fail with the following error:

================================================================================
⚠️  Run 5 (Pipeline 5) FAILED
================================================================================
Shard: 1/4
Error: ray::QueryProcessingActor.process_batch() (pid=2950, ip=192.168.10.10, actor_id=442f9bb3c78745a02ce0251901000000, repr=<rapidfireai.evals.actors.query_actor.QueryProcessingActor object at 0x7b0508aa8800>)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/rapidfireai/evals/actors/query_actor.py", line 324, in process_batch
    raise RuntimeError(
RuntimeError: Error processing batch: AttributeError: 'NoneType' object has no attribute 'get_context'
================================================================================
This run has been marked as FAILED. The experiment will continue with other runs.

To Reproduce

Steps to reproduce the behavior:

1. Run the Colab RAG notebook: https://colab.research.google.com/github/RapidFireAI/rapidfireai/blob/main/tutorial_notebooks/rag-contexteng/rf-colab-rag-fiqa-tutorial.ipynb
2. As the experiment is running, clone run 1, changing the chunk_size from 256 to 512.
3. See the cloned run fail and produce the error.

Expected Behavior

If the chunk_size can be changed, the run should successfully clone. If not, either chunk_size should not be presented as an option for cloning, or the server should reject the attempt and explain that chunk_size cannot be changed.

Environment

• Python version: 3.12
• RapidFire AI version: 0.14.0

Error Logs

================================================================================
⚠️  Run 5 (Pipeline 5) FAILED
================================================================================
Shard: 1/4
Error: ray::QueryProcessingActor.process_batch() (pid=2950, ip=192.168.10.10, actor_id=442f9bb3c78745a02ce0251901000000, repr=<rapidfireai.evals.actors.query_actor.QueryProcessingActor object at 0x7b0508aa8800>)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/rapidfireai/evals/actors/query_actor.py", line 324, in process_batch
    raise RuntimeError(
RuntimeError: Error processing batch: AttributeError: 'NoneType' object has no attribute 'get_context'
================================================================================
This run has been marked as FAILED. The experiment will continue with other runs.