Pipelines¶
The TransformersSharp.Pipeline class provides a base implementation for running pre-trained models using the Hugging Face Transformers library. It acts as a bridge between the Python-based Hugging Face pipelines and .NET applications.
What is a Pipeline?¶
A pipeline is a high-level abstraction that simplifies the process of using pre-trained models. Depending on the model, it handles tokenization, model inference, and decoding, allowing users to focus on their specific use cases without worrying about the underlying complexities.
Using the Pipeline Class¶
The Pipeline base class in TransformersSharp provides methods to run pre-trained models on input data.
Based on the task, the Pipeline class should be inherited, some examples include
- Text Generation Pipeline (
TextGenerationPipeline) - Text Classification Pipeline (`TextClassificationPipeline)
- Image Classification Pipeline(
ImageClassificationPipeline) - Object Detection Pipeline (
ObjectDetectionPipeline) - Text to Audio Pipeline (
TextToAudioPipeline)
Accessing the Tokenizer¶
The Pipeline class provides access to the associated tokenizer through the Tokenizer property. This allows users to preprocess inputs or decode outputs manually if needed.
Example: Accessing the Tokenizer¶
using TransformersSharp;
var pipeline = new Pipeline(pipelineObject); // Assume pipelineObject is initialized
var tokenizer = pipeline.Tokenizer;
var inputIds = tokenizer.Tokenize("How many helicopters can a human eat in one sitting?");
Console.WriteLine(string.Join(", ", inputIds.ToArray()));
Equivalent Python Code:
from transformers import pipeline
pipeline = pipeline("text-classification", model="facebook/opt-125m")
tokenizer = pipeline.tokenizer
input_ids = tokenizer("How many helicopters can a human eat in one sitting?", return_tensors="pt")["input_ids"]
print(input_ids.tolist())
Device Management¶
The Pipeline class automatically detects the device (CPU or GPU) being used by the underlying model. This information is accessible through the DeviceType property.
Example: Checking the Device Type¶
using TransformersSharp;
var pipeline = new Pipeline(pipelineObject); // Assume pipelineObject is initialized
Console.WriteLine($"Device Type: {pipeline.DeviceType}");
Equivalent Python Code: