-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathllms.txt
More file actions
105 lines (89 loc) · 4.85 KB
/
Copy pathllms.txt
File metadata and controls
105 lines (89 loc) · 4.85 KB
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
# state_collapser
`state_collapser` is a pre-alpha research package for constructing quotient
tower structure over discovered transition systems and exposing tower-aware
runtime and training surfaces.
Use this file as a compact routing map for agents, retrieval systems, and
humans. It is documentation, not a claim that LLMs or search systems will index
the project in any particular way.
## Start Here
- `README.md`: repository overview, positioning, installation, quick starts,
current maturity, and links to the main documentation.
- `EVALUATION.md`: current evaluation philosophy, example environments,
runnable probes, and the difference between smoke evaluation and serious
benchmarking.
- `CONTRIBUTING.md`: contributor workflow, current TODOs, release posture, and
repository layout.
- `SECURITY.md`: security policy and vulnerability-reporting guidance.
- `CITATION.cff`: citation metadata for the package.
- `CHANGELOG.md`: release history and current unreleased changes.
## Engineer Usage
- `docs/usage/01_001_what_state_collapser_is.md`: the shortest engineer-facing
explanation of what the package is and is not.
- `docs/usage/01_002_tower_runtime_mental_model.md`: tier direction,
partition-tower mental model, and local query vocabulary.
- `docs/usage/01_003_training_surface_quickstart.md`: how to use the current
training surfaces without hiding the training loop.
- `docs/usage/01_004_fiber_conditioned_training.md`: the current
fiber-conditioned training spine: freeze coarse behavior, train/lift in the
finer fiber.
- `docs/usage/01_005_using_your_own_training_loop.md`: guidance for engineers
bringing their own learner or training loop.
- `docs/usage/01_006_gymnasium_integration.md`: current Gymnasium integration
posture.
- `docs/usage/01_007_glossary.md`: project vocabulary.
- `docs/usage/01_008_common_misunderstandings.md`: common incorrect readings of
the package.
- `docs/usage/01_009_downstream_applications.md`: HGraphML and non-RL graph
dataflow context.
- `docs/usage/01_010_tensorization_boundary.md`: backend-independent
linearization, benchmark-mode reporting, and optional Torch batches.
## API Notes
- `docs/public_api.md`: public API policy and stability expectations.
- `docs/package_usage.md`: package usage routing and maturity notes.
- `docs/api_notes/partition_tower.md`: partition-tower runtime surface.
- `docs/api_notes/training_inputs_and_transitions.md`: training input and
transition records.
- `docs/api_notes/frozen_quotient_behavior.md`: frozen coarse-tier behavior.
- `docs/api_notes/path_fiber.md`: lift fibers over coarse decisions.
- `docs/api_notes/fiber_conditioned_stage.md`: adjacent-tier training stage
surface.
- `docs/api_notes/tensorization_boundary.md`: provisional tensorization API
names and benchmark labels.
## Research And Design
- `docs/design/logHRL.pdf`: companion research paper draft motivating
quotient-tower structure and logarithmic speed-up.
- `docs/design/logHRL.tex`: TeX source for the companion research paper draft.
- `docs/design/log_tropical_geometry/01_001_log_tropical_geometry_and_quotient_tower_discussion.md`:
research note on the relationship between quotient towers, scale
filtrations, log/tropical geometry, and semiring-valued dataflow.
- `docs/design/RL_framework_maturity/`: discussion, blueprints, and
implementation records for the fiber-conditioned training spine.
- `docs/design/Young_tableaux_refactor/`: design and implementation records for
the partition-table / Young-tableaux-style tower runtime.
- `docs/design/tensorization/`: design, implementation, and HGraphML bridge
notes for the first tensorization boundary.
- `docs/design/public_release_security_audit/`: public release, security, and
discoverability audit trail.
- `docs/engineer_continuity/`: continuity reports recording major engineering
phases and Project Owner attributions.
## Evaluation And Benchmarks
- `src/state_collapser/examples/`: example environments and runnable training
paths.
- `src/state_collapser/benchmarks/`: lightweight in-repo benchmark smoke
tooling.
- `tests/`: regression, partition runtime, training, example-environment, and
benchmark smoke tests.
- External serious benchmarking work is expected to live outside this package,
for example in the `big_boy_benchmarking` repository.
## Downstream Application
- `HGraphML`: downstream graph-ML application that uses `state_collapser`
partition towers over known graphs, runs message passing on coarse structure,
and lifts messages over node and edge fibers.
- Repository: `https://github.com/TYLERSFOSTER/HGraphML`
## Current Posture
- This is a public GitHub research release, not a PyPI release.
- The package is pre-alpha.
- The top-level stable import surface is intentionally small.
- Many useful internal modules are importable and tested, but not yet stable
public API.
- Serious benchmarking remains the major empirical-release need.