Falcon-accelerate-graph-vector-search

KO | EN

gitlite — search

#react #ai #golang #machine-learning #rust #macos #windows #awesome #cli #vue #awesome-list #security #ios #open-source

Falcon-accelerate-graph-vector-search

(★ 10)

No description available.

Falcon-accelerate-graph-vector-search 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# FPGA Accelerator For Graph-based Vector Search

This is a repository for our VLDB'25 paper [Fast Graph Vector Search via Hardware Acceleration and Delayed-Synchronization Traversal](https://arxiv.org/abs/2406.12385)

```
@article{jiang2024accelerating,
  title={Fast Graph Vector Search via Hardware Acceleration and Delayed-Synchronization Traversal},
  author={Jiang, Wenqi and Hu, Hang and Hoefler, Torsten and Alonso, Gustavo},
  journal={Proceedings of the VLDB Endowment},
  year={2025}
}
```

Software baseline: https://github.com/HangHu-sys/vector_search_baselines

## Related Projects

### ISCA'25: [RAGO: Systematic Performance Optimization for Retrieval-Augmented Generation Serving](https://arxiv.org/pdf/2503.14649).

System Performance Optimization for RAG.

Code: https://github.com/google/rago

```
@inproceedings{rago:isca:2025,
  title={RAGO: Systematic Performance Optimization for Retrieval-Augmented Generation Serving},
  author={Jiang, Wenqi and Subramanian, Suvinay and Graves, Cat and Alonso, Gustavo and Yazdanbakhsh, Amir and Dadu, Vidushi},
  booktitle = {Proceedings of the 52th Annual International Symposium on Computer Architecture}
  year={2025}
}
```

### KDD'25 [PipeRAG: Fast retrieval-augmented generation via adaptive pipeline parallelism](https://www.amazon.science/publications/piperag-fast-retrieval-augmented-generation-via-adaptive-pipeline-parallelism)

Efficient algorithm for iterative RAG.

Code: https://github.com/amazon-science/piperag

```
@article{jiang2025piperag,
  title={PipeRAG: Fast retrieval-augmented generation via adaptive pipeline parallelism},
  author={Jiang, Wenqi and Zhang, Shuai and Han, Boran and Wang, Jie and Wang, Yuyang Bernie and Kraska, Tim},
  journal={Proceedings of the 31th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
  year={2025}
}
```

### VLDB'25 [Chameleon: A Heterogeneous and Disaggregated Accelerator System for Retrieval-Augmented Language Models](https://arxiv.org/pdf/2310.09949)

Chameleon is a heterogeneous accelerator system for RAG serving. It prototypes FPGA-based accelerators for retrieval and GPU-based LLM inference.

Code: https://github.com/fpgasystems/Chameleon-RAG-Acceleration

```
@article{jiang2023chameleon,
  title={Chameleon: a heterogeneous and disaggregated accelerator system for retrieval-augmented language models},
  author={Jiang, Wenqi and Zeller, Marco and Waleffe, Roger and Hoefler, Torsten and Alonso, Gustavo},
  journal={Proceedings of the VLDB Endowment},
  year={2025}
}
```

### SC'23 [Co-design Hardware and Algorithm for Vector Search](https://arxiv.org/pdf/2306.11182)

Accelerating product-quantization-based vector search.

Code: https://github.com/WenqiJiang/SC-ANN-FPGA

```
@inproceedings{jiang2023co,
  title={Co-design hardware and algorithm for vector search},
  author={Jiang, Wenqi and Li, Shigang and Zhu, Yu and de Fine Licht, Johannes and He, Zhenhao and Shi, Runbin and Renggli, Cedric and Zhang, Shuai and Rekatsinas, Theodoros and Hoefler, Torsten and others},
  booktitle={Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis},
  pages={1--15},
  year={2023}
}
```

## All Bitstreams & Experiments

### Local Bitstreams & Experiments

#### Bitstreams

```
├── FPGA_multi_DDR
│   ├── FPGA_inter_query_v1.3_longer_FIFO_alt_PR
│   ├── FPGA_intra_query_v1.5_support_batching_longer_FIFO
```

Each bitstream may have different FIFO lengths that can pass P&R (already written in constant.hpp); Also double-check the connectivity.cfg file in case for different dimensionality, we need different P&R directives.

3 bitstreams for inter-query:

`FPGA_inter_query_v1.3_longer_FIFO_alt_PR`, with D=96, 128, 384 for Deep, SIFT, and SBERT; 4 channels

3 x 3 = 9 bitstreams for inter-query:

`FPGA_intra_query_v1.5_support_batching_longer_FIFO` , with D=96, 128, 384 for Deep, SIFT, and SBERT; and 1, 2, 4 channels for scalability tests

#### Experiments

See `perf_test_scripts/README.md` for the evaluation of throughput, latency, and recall for various settings.

### Networked bitstreams

#### Bitstreams

```
├── networked_FPGA
│   ├── kernel
│   │   └── user_krnl
│   │       ├── FPGA_inter_query_v1_3
│   │       ├── FPGA_intra_query_v1_5
```

For networked version, SBERT dataset would not work due to P&R problems with D=384.

2 bitstreams for inter-query:

`FPGA_inter_query_v1_3`, with D=96, 128; 4 channels

2 bitstreams for intra-query:

`FPGA_intra_query_v1_5`, with D=96, 128; 4 channels

#### Experiments

See `networked_FPGA/CPU_programs/README.md` for the evaluation of latency for various settings.

## Folder Organization

Here shows the most useful folders

```
├── FPGA_multi_DDR
│   ├── FPGA_inter_query_v1.3_longer_FIFO_alt_PR
│   ├── FPGA_intra_query_v1.5_support_batching_longer_FIFO
├── FPGA_single_DDR
├── networked_FPGA
│   ├── CPU_programs
│   ├── kernel
│   │   └── user_krnl
│   │       ├── FPGA_inter_query_v1_3
│   │       ├── FPGA_intra_query_v1_5
│   ├── host
│   │   ├── FPGA_inter_query_v1_3
│   │   ├── FPGA_intra_query_v1_5
├── perf_test_scripts
│   └── saved_df
├── plots
│   ├── images
├── test_dataflow_feedback
└── unit_tests
    ├── bloom_and_hash
    ├── bloom_fetch_compute
    ├── compute_PE
    ├── fetch_vectors
    └── priority_queue
```

`FPGA_multi_DDR`: multi-DDR version FPGA implementations

`networked_FPGA`: networked version of multi-DDR version FPGA implementations, with exactly the same kernel implemented in `FPGA_multi_DDR`

`perf_test_scripts`: all the plots used to measure the performance of local FPGAs

`plots`: all the plotting scripts

(unused) `FPGA_single_DDR`: single-DDR version FPGA implementations. Used only for developments.

(unused) `test_dataflow_feedback`: test the dataflow feedback loop behavior. Used only for developments.

(unused) `unit_tests`: benchmark the performance of different building blocks. Used only for developments.

Falcon-accelerate-graph-vector-search

Falcon-accelerate-graph-vector-search 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Falcon-accelerate-graph-vector-search

Falcon-accelerate-graph-vector-search 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README