Struct TextGenerationPipeline

pub struct TextGenerationPipeline { /* private fields */ }

A ready-to-use pipeline for generating text using a quantized model.

After building with TextGenerationPipelineBuilder, call generate_text(prompt, max_length) to produce text completions.

Example:

use transformers::pipelines::text_generation_pipeline::{TextGenerationPipelineBuilder, ModelOptions, Gemma3Size};

let pipeline = TextGenerationPipelineBuilder::new(
    ModelOptions::Gemma3(Gemma3Size::Size1B),
)
.temperature(0.7)
.build().unwrap();

let output = pipeline.generate_text("What is the meaning of life?", 5).unwrap();

println!("{}", output);

Implementations

impl TextGenerationPipeline

pub fn prompt_completion( &self, prompt: &str, max_length: usize, ) -> Result<String>

Examples found in repository

examples/text_generation_pipeline.rs (line 20)

fn main() -> Result<()> {
    println!("Building pipeline...");

    // 1. Create the pipeline, using the builder to configure the model
    let pipeline = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline built successfully.");

    // 2. Define prompt and max length
    let prompt = "Explain the concept of Large Language Models in simple terms.";
    let max_length = 250; // Maximum number of tokens to generate

    println!("Generating text for prompt: '{}'", prompt);

    // 3. Generate text
    let generated_text = pipeline.prompt_completion(prompt, max_length)?;

    println!("\n--- Generated Text ---");
    println!("{}", generated_text);
    println!("--- End of Text ---\n");

    let second_prompt = "Explain the fibonacci sequence in simple terms. /no_think";

    let generated_text = pipeline.prompt_completion(second_prompt, max_length)?;

    println!("\n--- Generated Text 2 ---");
    println!("{}", generated_text);
    println!("--- End of Text 2 ---");

    Ok(())
}

examples/test_multi.rs (line 74)

fn main() -> Result<()> {
    println!("Building pipeline 1...");
    // 1. Create the pipeline, using the builder to configure the model
    let pipeline_1 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 1 built successfully.");

    println!("Building pipeline 2...");
    let pipeline_2 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 2 built successfully.");

    println!("Building pipeline 3...");
    let pipeline_3 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 3 built successfully.");

    println!("Building pipeline 4...");
    let pipeline_4 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 4 built successfully.");

    println!("Building pipeline 5...");
    let pipeline_5 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 5 built successfully.");

    println!("Building pipeline 6...");
    let pipeline_6 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 6 built successfully.");

    println!("Building pipeline 7...");
    let pipeline_7 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 7 built successfully.");

    println!("Building pipeline 8...");
    let pipeline_8 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 8 built successfully.");

    println!("Building pipeline 9...");
    let pipeline_9 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 9 built successfully.");

    println!("Building pipeline 10...");
    let pipeline_10 = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size1_7B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline 10 built successfully.");

    /*

    Prompting all of them

    */

    let prompt = "Hello, world!";

    let response_1 = pipeline_1.prompt_completion(prompt, 10)?;
    println!("Response 1: {}", response_1);

    let response_2 = pipeline_2.prompt_completion(prompt, 10)?;
    println!("Response 2: {}", response_2);

    let response_3 = pipeline_3.prompt_completion(prompt, 10)?;
    println!("Response 3: {}", response_3);

    let response_4 = pipeline_4.prompt_completion(prompt, 10)?;
    println!("Response 4: {}", response_4);

    let response_5 = pipeline_5.prompt_completion(prompt, 10)?;
    println!("Response 5: {}", response_5);

    let response_6 = pipeline_6.prompt_completion(prompt, 10)?;
    println!("Response 6: {}", response_6);

    let response_7 = pipeline_7.prompt_completion(prompt, 10)?;
    println!("Response 7: {}", response_7);

    let response_8 = pipeline_8.prompt_completion(prompt, 10)?;
    println!("Response 8: {}", response_8);

    let response_9 = pipeline_9.prompt_completion(prompt, 10)?;
    println!("Response 9: {}", response_9);

    let response_10 = pipeline_10.prompt_completion(prompt, 100)?;
    println!("Response 10: {}", response_10);

    Ok(())
}

pub fn message_completion( &self, messages: Vec<Message>, max_length: usize, ) -> Result<String>

Examples found in repository

examples/text_generation_pipeline_with_messages.rs (line 32)

fn main() -> Result<()> {
    println!("Building pipeline...");

    // 1. Create the pipeline, using the builder to configure the model
    let pipeline = TextGenerationPipelineBuilder::new(ModelOptions::Qwen3(Qwen3Size::Size0_6B))
        .temperature(0.7)
        .build()?;
    println!("Pipeline built successfully.");

    // 2. Define messages and max length
    let mut messages = vec![
        Message::system("You are a helpful assistant."),
        Message::user("Explain the concept of Large Language Models in simple terms."),
    ];
    let max_length = 500; // Maximum number of tokens to generate

    // Get the last user message in the messages vector
    let prompt = messages
        .iter()
        .rev()
        .find(|message| message.role() == "user")
        .unwrap()
        .content();

    println!("Generating text for prompt: '{}'", prompt);

    // 3. Generate text
    let generated_text = pipeline.message_completion(messages.clone(), max_length)?;

    println!("\n--- Generated Text ---");
    println!("{}", generated_text);
    println!("--- End of Text ---\n");

    // Add the models response to the messages
    messages.push(Message::assistant(generated_text.as_str()));

    messages.push(Message::user(
        "Explain the fibonacci sequence in simple terms.",
    ));

    let generated_text = pipeline.message_completion(messages, max_length)?;

    println!("\n--- Generated Text 2 ---");
    println!("{}", generated_text);
    println!("--- End of Text 2 ---");

    Ok(())
}