TorchSharpCatalog.QuestionAnswer Method
Definition
Important
Some information relates to prerelease product that may be substantially modified before it’s released. Microsoft makes no warranties, express or implied, with respect to the information provided here.
Overloads
QuestionAnswer(MulticlassClassificationCatalog+MulticlassClassificationTrainers, QATrainer+Options) |
Fine tune a ROBERTA model for Question and Answer. The limit for any sentence is 512 tokens. Each word typically will map to a single token, and we automatically add 2 specical tokens (a start token and a separator token) so in general this limit will be 510 words for all sentences. |
QuestionAnswer(MulticlassClassificationCatalog+MulticlassClassificationTrainers, String, String, String, String, String, String, Int32, Int32, Int32, BertArchitecture, IDataView) |
Fine tune a ROBERTA model for Question and Answer. The limit for any sentence is 512 tokens. Each word typically will map to a single token, and we automatically add 2 specical tokens (a start token and a separator token) so in general this limit will be 510 words for all sentences. |
QuestionAnswer(MulticlassClassificationCatalog+MulticlassClassificationTrainers, QATrainer+Options)
Fine tune a ROBERTA model for Question and Answer. The limit for any sentence is 512 tokens. Each word typically will map to a single token, and we automatically add 2 specical tokens (a start token and a separator token) so in general this limit will be 510 words for all sentences.
public static Microsoft.ML.TorchSharp.Roberta.QATrainer QuestionAnswer (this Microsoft.ML.MulticlassClassificationCatalog.MulticlassClassificationTrainers catalog, Microsoft.ML.TorchSharp.Roberta.QATrainer.Options options);
static member QuestionAnswer : Microsoft.ML.MulticlassClassificationCatalog.MulticlassClassificationTrainers * Microsoft.ML.TorchSharp.Roberta.QATrainer.Options -> Microsoft.ML.TorchSharp.Roberta.QATrainer
<Extension()>
Public Function QuestionAnswer (catalog As MulticlassClassificationCatalog.MulticlassClassificationTrainers, options As QATrainer.Options) As QATrainer
Parameters
The transform's catalog.
- options
- QATrainer.Options
The options for QA.
Returns
Applies to
QuestionAnswer(MulticlassClassificationCatalog+MulticlassClassificationTrainers, String, String, String, String, String, String, Int32, Int32, Int32, BertArchitecture, IDataView)
Fine tune a ROBERTA model for Question and Answer. The limit for any sentence is 512 tokens. Each word typically will map to a single token, and we automatically add 2 specical tokens (a start token and a separator token) so in general this limit will be 510 words for all sentences.
public static Microsoft.ML.TorchSharp.Roberta.QATrainer QuestionAnswer (this Microsoft.ML.MulticlassClassificationCatalog.MulticlassClassificationTrainers catalog, string contextColumnName = "Context", string questionColumnName = "Question", string trainingAnswerColumnName = "TrainingAnswer", string answerIndexColumnName = "AnswerIndex", string predictedAnswerColumnName = "Answer", string scoreColumnName = "Score", int topK = 3, int batchSize = 4, int maxEpochs = 10, Microsoft.ML.TorchSharp.NasBert.BertArchitecture architecture = Microsoft.ML.TorchSharp.NasBert.BertArchitecture.Roberta, Microsoft.ML.IDataView validationSet = default);
static member QuestionAnswer : Microsoft.ML.MulticlassClassificationCatalog.MulticlassClassificationTrainers * string * string * string * string * string * string * int * int * int * Microsoft.ML.TorchSharp.NasBert.BertArchitecture * Microsoft.ML.IDataView -> Microsoft.ML.TorchSharp.Roberta.QATrainer
<Extension()>
Public Function QuestionAnswer (catalog As MulticlassClassificationCatalog.MulticlassClassificationTrainers, Optional contextColumnName As String = "Context", Optional questionColumnName As String = "Question", Optional trainingAnswerColumnName As String = "TrainingAnswer", Optional answerIndexColumnName As String = "AnswerIndex", Optional predictedAnswerColumnName As String = "Answer", Optional scoreColumnName As String = "Score", Optional topK As Integer = 3, Optional batchSize As Integer = 4, Optional maxEpochs As Integer = 10, Optional architecture As BertArchitecture = Microsoft.ML.TorchSharp.NasBert.BertArchitecture.Roberta, Optional validationSet As IDataView = Nothing) As QATrainer
Parameters
The transform's catalog.
- contextColumnName
- String
The context for the question.
- questionColumnName
- String
The question being asked.
- trainingAnswerColumnName
- String
The answer used to train the model.
- answerIndexColumnName
- String
The starting character index of that answer in the context.
- predictedAnswerColumnName
- String
The answer predicted by the model during inferencing.
- scoreColumnName
- String
The score of the predicted answers.
- topK
- Int32
How many top results you want back for a given question.
- batchSize
- Int32
Number of rows in the batch.
- maxEpochs
- Int32
Maximum number of times to loop through your training set.
- architecture
- BertArchitecture
Architecture for the model. Defaults to Roberta.
- validationSet
- IDataView
The validation set used while training to improve model quality.