Swiftide
Swiftide 是一个用于构建 LLM 应用程序的 Rust 库。它支持从简单的提示补全到快速、流式索引和查询管道，以及构建使用工具或调用其他代理的可组合代理。
高级功能
- 通用 LLM 任务的简单原语
- 流式索引和查询管道
- 可组合的代理和管道
- 模块化、可扩展的 API，抽象度最低
- 与流行的 LLM 和存储提供商集成
- 内置管道转换（或自带）
- 带有任务的图状工作流
- Langfuse 支持
安装
安装支持 Qdrant、OpenAI 和 Redis 的 Swiftide
cargo add swiftide --features=qdrant,openai,redis
请注意，Swiftide 默认是 barebones（裸机），因此您需要为您想要使用的集成启用功能。
索引示例（分步）
此示例使用 Swiftide 和 Qdrant 作为向量存储来索引 .rs 文件。
use swiftide::{
indexing,
indexing::LanguageModelWithBackOff,
indexing::loaders::FileLoader,
indexing::transformers::{ChunkCode, Embed, MetadataQACode},
integrations::{self, qdrant::Qdrant, redis::Redis},
};
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
tracing_subscriber::fmt::init();
// 1. Set up OpenAI client for embedding and prompt models
let openai_client = integrations::openai::OpenAI::builder()
.default_embed_model("text-embedding-3-small")
.default_prompt_model("gpt-3.5-turbo")
.build()?;
// 3. Set up Redis for caching which files/chunks are already processed
let redis_url = std::env::var("REDIS_URL")
.as_deref()
.unwrap_or("redis://:6379")
.to_owned();
indexing::Pipeline::from_loader(FileLoader::new(".").with_extensions(&["rs"]))
// 4. Skip files/chunks already indexed (cached in Redis)
.filter_cached(Redis::try_from_url(redis_url, "swiftide-examples")?)
// 5. Generate metadata Q&A for code chunks, using LLM
.then(MetadataQACode::new(openai_client.clone()))
// 6. Split code into chunks suitable for embedding
.then_chunk(ChunkCode::try_for_language_and_chunk_size("rust", 10..2048)?)
// 7. Embed code+metadata in batches
.then_in_batch(Embed::new(openai_client.clone()).with_batch_size(10))
// 8. Store results in a Qdrant collection
.then_store_with(
Qdrant::builder()
.batch_size(50)
.vector_size(1536)
.collection_name("swiftide-examples")
.build()?,
)
// 9. Run the pipeline asynchronously
.run()
.await?;
Ok(())
}
混合搜索示例
下面是使用 Qdrant 进行混合稠密/稀疏搜索的简化工作流。
use swiftide::{
indexing::{
self, EmbeddedField,
loaders::FileLoader,
transformers::{self, ChunkCode, MetadataQACode},
},
integrations::{fastembed::FastEmbed, openai, qdrant::Qdrant},
query::{self, answers, query_transformers, search_strategies::HybridSearch},
};
#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
tracing_subscriber::fmt::init();
// 1. Create fastembed (dense/sparse) clients
let batch_size = 64;
let fastembed_sparse = FastEmbed::try_default_sparse().unwrap().to_owned();
let fastembed = FastEmbed::try_default().unwrap().to_owned();
// 2. Use a compact OpenAI prompt model for metadata Q&A generation
let openai = openai::OpenAI::builder()
.default_prompt_model("gpt-4o-mini")
.build()
.unwrap();
// 3. Set up Qdrant for both dense and sparse vectors
let qdrant = Qdrant::builder()
.batch_size(batch_size)
.vector_size(384)
.with_vector(EmbeddedField::Combined)
.with_sparse_vector(EmbeddedField::Combined)
.collection_name("swiftide-hybrid-example")
.build()?;
indexing::Pipeline::from_loader(FileLoader::new("swiftide-core/").with_extensions(&["rs"]))
.then_chunk(ChunkCode::try_for_language_and_chunk_size("rust", 10..2048)?)
.then(MetadataQACode::from_client(openai.clone()).build().unwrap())
.then_in_batch(transformers::SparseEmbed::new(fastembed_sparse.clone()).with_batch_size(batch_size))
.then_in_batch(transformers::Embed::new(fastembed.clone()).with_batch_size(batch_size))
.then_store_with(qdrant.clone())
.run()
.await?;
// 4. Run a hybrid search pipeline
let openai = openai::OpenAI::builder()
.default_prompt_model("gpt-4o")
.build()
.unwrap();
let query_pipeline = query::Pipeline::from_search_strategy(
HybridSearch::default()
.with_top_n(20)
.with_top_k(20)
.to_owned(),
)
.then_transform_query(query_transformers::GenerateSubquestions::from_client(openai.clone()))
.then_transform_query(query_transformers::Embed::from_client(fastembed.clone()))
.then_transform_query(query_transformers::SparseEmbed::from_client(fastembed_sparse.clone()))
.then_retrieve(qdrant.clone())
.then_answer(answers::Simple::from_client(openai.clone()));
let answer = query_pipeline
.query("What are the different pipelines in Swiftide and how do they work?")
.await
.unwrap();
println!("{}", answer.answer());
}