Tunable text classification

Overview

Tunable estimators allow to fine-tune the underlying LLM for a classification task. Usually, tuning is performed directly in the cloud (e.g. OpenAI, Vertex), therefore it is not required to have a GPU on your local machine. However, be aware that tuning can be costly and time-consuming. We recommend to first try the in-context learning estimators, and only if they do not provide satisfactory results, to try the tunable estimators.

Example using GPT-3.5-Turbo-0613:

from skllm.models.gpt.classification.tunable import GPTClassifier

X, y = get_classification_dataset()
clf = GPTClassifier(n_epochs=1)
clf.fit(X,y)
clf.predict(X)

API Reference

The following API reference only lists the parameters needed for the initialization of the estimator. The remaining methods follow the syntax of a scikit-learn classifier.

GPTClassifier

from skllm.models.gpt.classification.tunable import GPTClassifier

Parameter	Type	Description
`base_model`	`str`	Base model to use, by default "gpt-3.5-turbo-0613".
`default_label`	`str`	Default label for failed prediction; if "Random" -> selects randomly based on class frequencies, by default "Random".
`key`	`Optional[str]`	Estimator-specific API key; if None, retrieved from the global config, by default None.
`org`	`Optional[str]`	Estimator-specific ORG key; if None, retrieved from the global config, by default None.
`n_epochs`	`Optional[int]`	Number of epochs; if None, determined automatically; by default None.
`custom_suffix`	`Optional[str]`	Custom suffix of the tuned model, used for naming purposes only, by default "skllm".
`prompt_template`	`Optional[str]`	Custom prompt template to use, by default None.

MultiLabelGPTClassifier

from skllm.models.gpt.classification.tunable import MultiLabelGPTClassifier

Parameter	Type	Description
`base_model`	`str`	Base model to use, by default "gpt-3.5-turbo-0613".
`default_label`	`str`	Default label for failed prediction; if "Random" -> selects randomly based on class frequencies, by default "Random".
`key`	`Optional[str]`	Estimator-specific API key; if None, retrieved from the global config, by default None.
`org`	`Optional[str]`	Estimator-specific ORG key; if None, retrieved from the global config, by default None.
`n_epochs`	`Optional[int]`	Number of epochs; if None, determined automatically; by default None.
`custom_suffix`	`Optional[str]`	Custom suffix of the tuned model, used for naming purposes only, by default "skllm".
`prompt_template`	`Optional[str]`	Custom prompt template to use, by default None.
`max_labels`	`Optional[int]`	Maximum labels per sample, by default 5.

VertexClassifier

from skllm.models.vertex.classification.tunable import VertexClassifier

Parameter	Type	Description
`base_model`	`str`	Base model to use, by default "text-bison@002".
`n_update_steps`	`int`	Number of epochs, by default 1.
`default_label`	`str`	Default label for failed prediction; if "Random" -> selects randomly based on class frequencies, by default "Random".