Model tuning
Overview
Prompt tuning is an efficient and low-cost way to adapt an AI foundation model to new downstream tasks without retraining the model and updating its weights. You can tune foundation models in IBM watsonx.ai programmatically by using the Python library.
Quick start
The TuneExperiment class is responsible for creating experiments and scheduling tunings. All experiment results are stored automatically in the user-specified IBM Cloud Object Storage for SaaS or in the cluster’s file system for Cloud Pak for Data. Then, the TuneExperiment feature can fetch the results and provide them directly to the user for further usage.
Import and configure PromptTuner
Replace {watsonx_ai_url}, {apikey}, {project_id}, and {space_id} with your information.
1from ibm_watsonx_ai import Credentials 2from ibm_watsonx_ai.experiment import TuneExperiment 3 4credentials = Credentials( 5 url = "{watsonx_ai_url}", 6 apikey = "{apikey}", 7) 8 9experiment = TuneExperiment( 10 credentials=Credentials(...), 11 project_id = "{project_id}" 12 space_id="{space_id}") 13 14prompt_tuner = experiment.prompt_tuner( 15 name="prompt tuning name", 16 task_id=experiment.Tasks.CLASSIFICATION, 17 base_model=ModelTypes.FLAN_T5_XL, 18 accumulate_steps=32, 19 batch_size=16, 20 learning_rate=0.2, 21 max_input_tokens=256, 22 max_output_tokens=2, 23 num_epochs=6, 24 tuning_type=experiment.PromptTuningTypes.PT, 25 verbalizer="Extract the satisfaction from the comment. Return simple '1' for satisfied customer or '0' for unsatisfied. Input: {{input}} Output: ", 26 auto_update_model=True 27)
Run prompt tuning
1from ibm_watsonx_ai.helpers import DataConnection, ContainerLocation, S3Location 2 3tuning_details = prompt_tuner.run( 4 training_data_references=[DataConnection( 5 connection_asset_id=connection_id, 6 location=S3Location( 7 bucket='prompt_tuning_data', 8 path='pt_train_data.json' 9 ) 10 )], 11 background_mode=False) 12 13# OR 14 15tuning_details = prompt_tuner.run( 16 training_data_references=[DataConnection( 17 data_asset_id='5d99c11a-2060-4ef6-83d5-dc593c6455e2') 18 ], 19 background_mode=True) 20 21# OR 22 23tuning_details = prompt_tuner.run( 24 training_data_references=[DataConnection( 25 location=ContainerLocation("path_to_file.json")) 26 ], 27 background_mode=True)
The watsonx.ai Python library has methods and helper classes for tuning foundation models. For more information about the library, see Prompt tuning.