Different Outputs Between Local Nuextract Model and Hugging Face Playground
I'm encountering an issue where the output from the Nuextract model differs when running it locally versus using it on the Hugging Face Playground.
Local Model: The model captures some irrelevant information that isn’t required.
Hugging Face Playground: The output is accurate and as expected.
I'm wondering if there are specific parameters or configurations that need to be adjusted locally to match the behavior of the Playground.
Has anyone faced a similar issue or can suggest what settings might be causing this difference?
Are you using a temperature of 0.0?
Yes
yes you are not alone
May We know which default params are used?
thank you in advance
Ok, it seems the model running on the space was actually using sampling which is likely the issue (It is now updated). Were you finding the space to be consistently giving better results?