llama_index/llama-index-core/llama

At a glance

The community member is experiencing an issue where the textNode they are using is getting truncated with a "..." due to a specific code path in the llama_index library. This is causing issues with their response synthesis, as the full text response is not available. The community members discuss potential workarounds, such as using node.text or node.get_content() instead of the default behavior, and creating a custom component that returns strings instead of nodes. The community members also suggest reporting the issue and potentially submitting a pull request to fix the bug.

Useful resources

nnspielbe

Hi there, I have an issue with the textNode I am using getting truncated with a "..." because of https://github.com/run-llama/llama_index/blob/main/llama-index-core/llama_index/core/schema.py#L329-L336

Where my Text node is being truncated so my response synthesis cannot use the full text response because of the truncation. Has anyone hit this issue or am I doing something wrong? Thanks!

13 comments

LLogan M

What code are you running?

All the internal code should be using node.text or node.get_content()

nnspielbe

sql_retriever = SQLRetriever(sql_database)

The output of this is calling
str method of the text node which incorrectly is passing into the context when I was debugging

nnspielbe

in query.py validate and convert stringable: https://github.com/run-llama/llama_index/blob/main/llama-index-core/llama_index/core/base/query_pipeline/query.py#L61-L62

bc node with score is a stringable input-- I see this getting called

LLogan M

Interesting, seems like a bug

LLogan M

Workaround would be a custom component that returns strings instead of nodes

nnspielbe

yeah- tried this and it works (overwrote locally as well also works)

nnspielbe

@Logan M should I report an issue or?

LLogan M

yea probably worth it. Although if you can think of a fix, a PR would be even better ❤️