Copilot Context Issue in Fabric Notebook

[Anonymous]

Hi all,

In Fabric, I'm trying to use Copilot in a Notebook. I have a DataFrame df that I created by reading a table from my lakehouse. When I try to run the following code asking Copilot to summarize the data in df:

%describe
df

 I receive the following error:

RuntimeError: StructuredContext.optimize() was called with a mandatory context that was too large to fit in the max_prompt_tokens limit. The mandatory context was 3626 tokens, but the limit was 3346 tokens.

It's clear an input exceeds the maximum token length. I have nothing else loaded into memory other than df, which has a shape of (137923, 13) though the error still occurs if I try with only 1,000 rows instead. I am not aware of a way to control the context that Copilot includes before passing to gpt-35-turbo-0125. Other magic commands such as %%chat and %%code work normally without issue.

Has anyone encountered this or have any tips or things to try? Also, if there's a better place to post this, please let me know. Thanks!

[Anonymous]

Actually it's not just %describe that's having the issue, I'm getting the same error if I try:

%%chat
Analyze the data in df

RuntimeError: StructuredContext.optimize() was called with a mandatory context that was too large to fit in the max_prompt_tokens limit. The mandatory context was 3649 tokens, but the limit was 3346 tokens.

[lbendlin]

Check this video  Microsoft Fabric Notebook Copilot Tutorial - YouTube  although it stops just short of the %%describe example.

[lbendlin]

you're not accidentally omitting the second percent sign?

%%describe df

[Anonymous · Accepted solution]

RESOLVED

The issue was that the Lakehouse I was loading a table from had too many tables (only 20 tables, each with ~100,000 rows).

For anyone else encountering this issue, try:

1. Creating a new larger Spark pool (they're currently medium by default, x-large fixed my issue). See your Workspace settings.
2. Reducing the size of your Lakehouse by deleting tables (not an option for my case)
3. Creating a new Lakehouse to reference with the table you're trying to load (worked for me when my new Lakehouse was in a different Workspace)

[Anonymous]

Thanks for the quick response! No, I tried that first but for some reason that magic command prefers a single %:

UsageError: Cell magic `%%describe` not found (But line magic `%describe` exists, did you mean that instead?).