blob: 7bd7a72d8194a17e246f261e49fb915acbbbe083 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
|
# HAG: Hopfield-Augmented Generation
A retrieval-augmented generation system that replaces standard vector similarity search (FAISS top-k) with **iterative Modern Hopfield Network retrieval**.
## Quick Start
```bash
pip install -e ".[dev]"
pytest
```
## Core Idea
Standard RAG retrieves passages once via FAISS similarity. HAG iteratively refines the query in embedding space using Modern Continuous Hopfield Network dynamics:
```
q_{t+1} = M * softmax(beta * M^T * q_t)
```
This enables implicit multi-hop reasoning through energy minimization with convergence guarantees.
## Project Structure
- `hag/hopfield.py` — Core Hopfield retrieval module
- `hag/memory_bank.py` — Passage embedding storage
- `hag/retriever_hopfield.py` — Hopfield-based retriever
- `hag/retriever_faiss.py` — FAISS baseline retriever
- `hag/pipeline.py` — End-to-end RAG/HAG pipeline
- `hag/metrics.py` — Evaluation metrics (EM, F1, retrieval recall)
- `hag/energy.py` — Energy analysis utilities
- `scripts/` — Evaluation and analysis scripts
- `tests/` — Comprehensive test suite
## Running Tests
```bash
pytest tests/ -v
```
All tests run on CPU with synthetic data — no GPU or model downloads required.
|
