feat: Add reranking functionality to Jina client

- Added a new `rerank` function to the Jina client for reranking documents.
- Added a new `RerankParams` struct to define parameters for reranking.
- Added unit tests for the new `rerank` function.
- Updated the example script to demonstrate reranking.
- Improved error handling and added more comprehensive logging.
This commit is contained in:
Mahmoud Emad
2025-03-11 19:27:01 +02:00
parent 7965883744
commit 0e1836c5d0
5 changed files with 131 additions and 32 deletions

View File

@@ -11,3 +11,12 @@ embeddings := jina_client.create_embeddings(
) or { panic('Error while creating embeddings: ${err}') }
println('Created embeddings: ${embeddings}')
rerank_result := jina_client.rerank(
model: .reranker_v2_base_multilingual
query: 'skincare products'
documents: ['Product A', 'Product B', 'Product C']
top_n: 2
) or { panic('Error while reranking: ${err}') }
println('Rerank result: ${rerank_result}')

View File

@@ -47,6 +47,44 @@ pub fn (mut j Jina) create_embeddings(params CreateEmbeddingParams) !ModelEmbedd
return parse_model_embedding_output(response)!
}
@[params]
pub struct RerankParams {
pub mut:
model JinaRerankModel @[required]
query string @[required]
documents []string @[required]
top_n ?int // Optional: Number of top results to return
return_documents ?bool // Optional: Flag to determine if the documents should be returned
}
// Rerank documents based on a query
pub fn (mut j Jina) rerank(params RerankParams) !RankingOutput {
mut rank_input := RerankInput{
model: params.model.to_string()
query: params.query
documents: params.documents
}
if v := params.top_n {
rank_input.top_n = v
}
if v := params.return_documents {
rank_input.return_documents = v
}
req := httpconnection.Request{
method: .post
prefix: 'v1/rerank'
dataformat: .json
data: json.encode(rank_input)
}
mut httpclient := j.httpclient()!
response := httpclient.post_json_str(req)!
return json.decode(RankingOutput, response)!
}
// // Create embeddings with a TextDoc input
// pub fn (mut j Jina) create_embeddings_with_docs(args TextEmbeddingInput) !ModelEmbeddingOutput {

View File

@@ -1,11 +1,14 @@
module jina
import time
fn setup_client() !&Jina {
mut client := get()!
return client
}
fn test_create_embeddings() {
time.sleep(1 * time.second)
mut client := setup_client()!
embeddings := client.create_embeddings(
input: ['Hello', 'World']
@@ -17,3 +20,17 @@ fn test_create_embeddings() {
assert embeddings.object == 'list' // Check the object type
assert embeddings.model == 'jina-embeddings-v3'
}
fn test_rerank() {
time.sleep(1 * time.second)
mut client := setup_client()!
rerank_result := client.rerank(
model: .reranker_v2_base_multilingual
query: 'skincare products'
documents: ['Product A', 'Product B', 'Product C']
top_n: 2
) or { panic('Error while reranking: ${err}') }
assert rerank_result.results.len == 2
assert rerank_result.model == 'jina-reranker-v2-base-multilingual'
}

View File

@@ -2,38 +2,6 @@ module jina
import json
// RankAPIInput represents the input for reranking requests
// model:
// jina-reranker-v2-base-multilingual, 278M
// jina-reranker-v1-base-en, 137M
// jina-reranker-v1-tiny-en, 33M
// jina-reranker-v1-turbo-en, 38M
// jina-colbert-v1-en, 137M
pub struct RankAPIInputRAW {
pub mut:
model string @[required]
query string @[required]
documents []string @[required]
top_n int // Optional: Number of top results to return
}
// RankingOutput represents the response from reranking requests
pub struct RankingOutput {
pub mut:
model string
results []RankResult
usage Usage
object string
}
// RankResult represents a single reranking result
pub struct RankResult {
pub mut:
document string
index int
relevance_score f64
}
// ClassificationAPIInput represents the input for classification requests
pub struct ClassificationAPIInput {
pub mut:

View File

@@ -0,0 +1,67 @@
module jina
import json
pub enum JinaRerankModel {
reranker_v2_base_multilingual // 278M
reranker_v1_base_en // 137M
reranker_v1_tiny_en // 33M
reranker_v1_turbo_en // 38M
colbert_v1_en // 137M
}
// RankAPIInput represents the input for reranking requests
pub struct RerankInput {
pub mut:
model string @[required] // Model name
query string @[required] // Query text
documents []string @[required] // Document texts
top_n ?int // Optional: Number of top results to return
return_documents ?bool // Optional: Flag to determine if the documents should be returned
}
// RankingOutput represents the response from reranking requests
pub struct RankingOutput {
pub mut:
model string
results []RankResult
usage Usage
object string
}
// RankResult represents a single reranking result
pub struct RankResult {
pub mut:
document RankDocument
index int
relevance_score f64
}
// RankDocument represents a single document for reranking
pub struct RankDocument {
pub mut:
text string
}
// to_string converts a JinaRerankModel enum to its string representation as expected by the Jina API
pub fn (m JinaRerankModel) to_string() string {
return match m {
.reranker_v2_base_multilingual { 'jina-reranker-v2-base-multilingual' }
.reranker_v1_base_en { 'jina-reranker-v1-base-en' }
.reranker_v1_tiny_en { 'jina-reranker-v1-tiny-en' }
.reranker_v1_turbo_en { 'jina-reranker-v1-turbo-en' }
.colbert_v1_en { 'jina-colbert-v1-en' }
}
}
// from_string converts a string to a JinaRerankModel enum, returning an error if the string is invalid
pub fn jina_rerank_model_from_string(s string) !JinaRerankModel {
return match s {
'jina-reranker-v2-base-multilingual' { JinaRerankModel.reranker_v2_base_multilingual }
'jina-reranker-v1-base-en' { JinaRerankModel.reranker_v1_base_en }
'jina-reranker-v1-tiny-en' { JinaRerankModel.reranker_v1_tiny_en }
'jina-reranker-v1-turbo-en' { JinaRerankModel.reranker_v1_turbo_en }
'jina-colbert-v1-en' { JinaRerankModel.colbert_v1_en }
else { error('Invalid JinaRerankModel string: ${s}') }
}
}