| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Example of using the OpenAI entrypoint's rerank API which is compatible with |
| 4 | the Cohere SDK: https://github.com/cohere-ai/cohere-python |
| 5 | Note that `pip install cohere` is needed to run this example. |
| 6 | |
| 7 | run: vllm serve BAAI/bge-reranker-base |
| 8 | """ |
| 9 | |
| 10 | import cohere |
| 11 | from cohere import Client, ClientV2 |
| 12 | |
| 13 | model = "BAAI/bge-reranker-base" |
| 14 | |
| 15 | query = "What is the capital of France?" |
| 16 | |
| 17 | documents = [ |
| 18 | "The capital of France is Paris", |
| 19 | "Reranking is fun!", |
| 20 | "vLLM is an open-source framework for fast AI serving", |
| 21 | ] |
| 22 | |
| 23 | |
| 24 | def cohere_rerank( |
| 25 | client: Client | ClientV2, model: str, query: str, documents: list[str] |
| 26 | ) -> dict: |
| 27 | return client.rerank(model=model, query=query, documents=documents) |
| 28 | |
| 29 | |
| 30 | def main(): |
| 31 | # cohere v1 client |
| 32 | cohere_v1 = cohere.Client(base_url="http://localhost:8000", api_key="sk-fake-key") |
| 33 | rerank_v1_result = cohere_rerank(cohere_v1, model, query, documents) |
| 34 | print("-" * 50) |
| 35 | print("rerank_v1_result:\n", rerank_v1_result) |
| 36 | print("-" * 50) |
| 37 | |
| 38 | # or the v2 |
| 39 | cohere_v2 = cohere.ClientV2("sk-fake-key", base_url="http://localhost:8000") |
| 40 | rerank_v2_result = cohere_rerank(cohere_v2, model, query, documents) |
| 41 | print("rerank_v2_result:\n", rerank_v2_result) |
| 42 | print("-" * 50) |
| 43 | |
| 44 | |
| 45 | if __name__ == "__main__": |
| 46 | main() |