Number of created nodes is changing #13124
Replies: 1 comment
-
The discrepancy in the number of nodes created when running your function in a Jupyter notebook versus Visual Studio Code, especially with Recursive Retrieval in LlamaIndex, can indeed occur due to differences in execution environments. These differences can affect how memory, parallel tasks, and temporary files are managed, leading to variations in the node creation process. The complexity of recursive retrieval, which involves managing sub-nodes of varying sizes and their interconnections, is sensitive to these environmental differences. Furthermore, the interaction with underlying libraries, system resources, and external services like vector storage can also contribute to the observed discrepancies. It's important to consider these factors when evaluating the consistency of node creation across different platforms. For a deeper understanding and potential solutions, reviewing the documentation on Recursive Retrieval Nodes and comparing Auto vs. Recursive Retriever methods might provide insights into managing or mitigating these discrepancies.
|
Beta Was this translation helpful? Give feedback.
-
I am testing the function that creates the nodes from a document. While using that function in a Jupyter notebook, it retrieves 82 nodes in total, while testing it in vscode for the final implementation, it retrieves 92.
The environments use the same versions of python and LlamaIndex, and the code is exactly the same.
Has this happened to any of you? Should I be worried about it not having the same number of nodes?
Pd: I'm using Recursive Retrieval and the difference gets bigger as the size of the sub nodes gets lower.
Beta Was this translation helpful? Give feedback.
All reactions