turbo-graph/turbo-graph-python/examples/graph_memory_rag.py at main · bigmacfive/turbo-graph · GitHub

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
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
"""Minimal graph-memory RAG workflow.

Run after installing the wheel:

    python turbo-graph-python/examples/graph_memory_rag.py

The example models the part turbovec leaves to application code: combine a
local graph neighborhood with tags, source ACLs, time windows, and upstream
candidate ids before vector search.
"""

from __future__ import annotations

import json
from pathlib import Path
from tempfile import TemporaryDirectory

import numpy as np

from turbo_graph import GraphMemoryIndex


DIM = 8


def unit(axis: int) -> np.ndarray:
    vector = np.zeros(DIM, dtype=np.float32)
    vector[axis] = 1.0
    return vector


def build_memory() -> tuple[GraphMemoryIndex, np.ndarray]:
    vectors = np.ascontiguousarray(
        np.vstack([unit(0), unit(1), unit(2), unit(3), unit(4)])
    )
    records = [
        {
            "id": 100,
            "title": "Architecture root",
            "tags": ["architecture", "root"],
            "source": "docs",
            "timestamp_ms": 1_700_000_000_000,
        },
        {
            "id": 110,
            "title": "Retrieval cache design",
            "tags": ["architecture", "cache"],
            "source": "docs",
            "timestamp_ms": 1_700_000_100_000,
        },
        {
            "id": 120,
            "title": "Ops incident",
            "tags": ["ops"],
            "source": "tickets",
            "timestamp_ms": 1_700_000_200_000,
        },
        {
            "id": 130,
            "title": "Release checklist",
            "tags": ["release"],
            "source": "docs",
            "timestamp_ms": 1_700_000_300_000,
        },
        {
            "id": 140,
            "title": "Archived idea",
            "tags": ["archive"],
            "source": "archive",
            "timestamp_ms": 1_700_000_400_000,
        },
    ]

    memory = GraphMemoryIndex(dim=DIM, bit_width=4)
    memory.add_records(vectors, records)
    memory.link_bidirectional(100, 110, 1.0)
    memory.link_directed(110, 120, 0.7)
    memory.link_directed(100, 130, 0.4)
    return memory, vectors


def main() -> None:
    memory, vectors = build_memory()

    upstream_candidates = [110, 120, 999, 110]
    hits = memory.search(
        vectors[1],
        k=5,
        seeds=[100],
        max_hops=2,
        required_tags=["architecture"],
        allowed_sources=["docs"],
        start_ms=1_700_000_050_000,
        end_ms=1_700_000_200_000,
        candidate_ids=upstream_candidates,
    )

    report = memory.explain(
        vectors[1],
        k=5,
        seeds=[100],
        preset="balanced",
        required_tags=["architecture"],
        allowed_sources=["docs"],
        candidate_ids=upstream_candidates,
    )

    batch_hits = memory.search_batch(
        np.ascontiguousarray(np.vstack([vectors[1], vectors[2]])),
        k=5,
        seeds=[100],
        max_hops=2,
        required_tags=["architecture"],
        candidate_ids=upstream_candidates,
    )

    before_trim = memory.cache_stats()
    budget = memory.trim_caches_for_preset("low_latency")
    after_trim = memory.cache_stats()

    with TemporaryDirectory() as tmp:
        index_path = Path(tmp) / "memory.tv"
        graph_path = Path(tmp) / "memory.tvgm"
        memory.write(str(index_path), str(graph_path))
        loaded = GraphMemoryIndex.load(str(index_path), str(graph_path))
        assert loaded.record(110)["title"] == "Retrieval cache design"

    assert [hit["id"] for hit in hits] == [110]
    assert [len(row) for row in batch_hits] == [1, 1]
    assert report["plan"]["candidate_input_ids"] == 4
    assert report["plan"]["candidate_missing_ids"] == 1
    assert report["plan"]["candidate_duplicate_ids"] == 1
    assert before_trim["total_entries"] >= after_trim["total_entries"]

    print(
        json.dumps(
            {
                "hits": hits,
                "batch_hit_ids": [[hit["id"] for hit in row] for row in batch_hits],
                "candidate_plan": {
                    key: report["plan"][key]
                    for key in (
                        "selected_slots",
                        "candidate_input_ids",
                        "candidate_slots",
                        "candidate_missing_ids",
                        "candidate_duplicate_ids",
                    )
                },
                "cache_budget_total_entries": budget["total_entries"],
                "cache_entries_after_trim": after_trim["total_entries"],
            },
            indent=2,
            sort_keys=True,
        )
    )


if __name__ == "__main__":
    main()