Table 10 Best performing hyperparameters for all approaches

From: Detecting stigmatizing language in clinical notes with large language models for addiction care

Approach

Chunk Size (Tokens)

Chunk Overlap (Tokens)

Scoring

Num. Context Entries

Context Source

Supervised-Fine Tuning (SFT)

N/A

N/A

N/A

N/A

N/A

In-context

1000

100

N/A

N/A

5

Retrieval Augmented Prompt (RAG)

1000

100

dot

5

13

Zero-Shot Prompt

1000

100

N/A

N/A

N/A

Manual

N/A

N/A

N/A

N/A

N/A

  1. These hyperparameter values were selected for using the results from the validation dataset where all hyperparameter sets were scored on this subset of the MIMIC-III17 data.
Back to article page