Optitransfer commited on
Commit
ae0f017
·
verified ·
1 Parent(s): bbce1fc

rebrand: convergent intelligence infrastructure

Browse files
Files changed (1) hide show
  1. README.md +91 -47
README.md CHANGED
@@ -1,86 +1,130 @@
1
  ---
2
  title: Optitransfer
3
- emoji: 🔀
4
  sdk: static
5
  pinned: true
6
- colorFrom: indigo
7
- colorTo: blue
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8
  ---
9
 
 
 
10
  # Optitransfer
11
 
12
- Convergent intelligence infrastructure for distributed AI systems. Patent-pending.
 
 
 
 
 
 
 
 
 
 
 
 
 
13
 
14
  ---
15
 
16
- ## What We Build
 
 
 
 
17
 
18
- **[crdt-merge](https://github.com/mgillr/crdt-merge)** the open-source library that makes AI model merging deterministic, conflict-free, and mathematically guaranteed to converge. Any number of models, any merge order, provably correct result. Every time.
19
 
20
- crdt-merge applies Conflict-Free Replicated Data Type theory to neural network weight spaces. The result: models that merge like distributed databases — without coordination, without data loss, without conflicts.
 
 
21
 
22
- **Patent Pending:** UK Application GB 2607132.4
23
 
24
  ---
25
 
26
- ## Core Technology
 
 
 
 
 
 
 
 
27
 
28
- | Capability | What It Does |
29
- |---|---|
30
- | **26 Merge Strategies** | From simple averaging to adaptive trust-weighted convergence |
31
- | **Merkle Provenance** | Every weight, every merge, cryptographically tracked |
32
- | **Causal Ordering** | Vector clocks ensure correct merge sequence across distributed nodes |
33
- | **Byzantine Tolerance** | Resilient against adversarial or faulty participants |
34
- | **Delta Compression** | Ship only what changed — not entire models |
35
- | **Format Agnostic** | PyTorch, TensorFlow, SafeTensors, GGUF, NumPy |
36
 
37
  ---
38
 
39
- ## Live Demos
40
 
41
- | Space | Description |
42
- |---|---|
43
- | **[crdt-merge](https://huggingface.co/spaces/Optitransfer/crdt-merge)** | Flagship — merge real models with full provenance and compliance |
44
- | **[Convergence Lab](https://huggingface.co/spaces/Optitransfer/convergence-lab)** | Test all 26 strategies, visualise convergence, run experiments |
45
- | **[Data Playground](https://huggingface.co/spaces/Optitransfer/crdt-merge-data)** | CRDT merging applied to tabular data |
46
- | **[Federation](https://huggingface.co/spaces/Optitransfer/crdt-merge-federation)** | Multi-node gossip convergence simulation |
47
 
48
  ---
49
 
50
- ## Getting Started
51
 
52
- ```bash
53
- pip install crdt-merge
54
- ```
 
 
 
 
 
55
 
56
- ```python
57
- from crdt_merge import CRDTMergeState
58
 
59
- state = CRDTMergeState()
60
- state.load_model("model_a.safetensors")
61
- state.load_model("model_b.safetensors")
62
- result = state.merge(strategy="dare_ties")
63
- result.save("merged.safetensors")
64
- ```
 
 
 
 
 
65
 
66
  ---
67
 
68
  ## Data Products
69
 
70
- Optitransfer also produces compliance-ready, quality-scored web corpora for AI training.
71
 
72
- | Dataset | Records | Access |
73
- |---|---|---|
74
- | [Swiss Web Premium (A+)](https://huggingface.co/datasets/OptiTransferData/swiss-web-premium-ch) | 110,491 | [Sample](https://huggingface.co/datasets/OptiTransferData/swiss-web-premium-ch) \| [Full](https://huggingface.co/datasets/OptiTransferData/swiss-web-premium-ch-full) |
75
-
76
- EU AI Act compliant. PII-redacted. SHA256-verified. Multilingual (DE, FR, EN, IT + 25 languages).
77
 
78
  ---
79
 
80
- ## Contact
 
 
81
 
82
- - **GitHub:** [github.com/mgillr/crdt-merge](https://github.com/mgillr/crdt-merge)
83
- - **PyPI:** [pypi.org/project/crdt-merge](https://pypi.org/project/crdt-merge/)
84
- - **Email:** [data@optitransfer.ch](mailto:data@optitransfer.ch)
85
- - **Location:** Switzerland
86
 
 
 
1
  ---
2
  title: Optitransfer
3
+ emoji: 🔬
4
  sdk: static
5
  pinned: true
6
+ colorFrom: gray
7
+ colorTo: gray
8
+ tags:
9
+ - model-merging
10
+ - crdt
11
+ - distributed-systems
12
+ - federated-learning
13
+ - convergence
14
+ - deterministic
15
+ - conflict-free
16
+ - byzantine-fault-tolerance
17
+ - agent-memory
18
+ - provenance
19
+ - safetensors
20
+ - pytorch
21
+ - neural-networks
22
+ - multi-agent
23
  ---
24
 
25
+ <div align="center">
26
+
27
  # Optitransfer
28
 
29
+ **Convergent intelligence infrastructure.**
30
+
31
+ *Building the mathematical layer where distributed AI systems merge, verify, and evolve -- deterministically.*
32
+
33
+ ---
34
+
35
+ [![PyPI](https://img.shields.io/pypi/v/crdt-merge?label=PyPI&color=blue)](https://pypi.org/project/crdt-merge/)
36
+ [![Downloads](https://img.shields.io/pypi/dm/crdt-merge?label=Downloads&color=brightgreen)](https://pepy.tech/projects/crdt-merge)
37
+ [![Tests](https://img.shields.io/badge/tests-4%2C498%20passing-brightgreen)](https://github.com/mgillr/crdt-merge)
38
+ [![CRDT Compliance](https://img.shields.io/badge/CRDT%20compliance-26%2F26-blue)](https://github.com/mgillr/crdt-merge/blob/main/docs/CRDT_ARCHITECTURE.md)
39
+ [![License](https://img.shields.io/badge/license-BSL%201.1%20→%20Apache%202.0-orange)](https://github.com/mgillr/crdt-merge/blob/main/LICENSE)
40
+ [![Patent](https://img.shields.io/badge/UK%20Patent-pending-lightgrey)](https://github.com/mgillr/crdt-merge/blob/main/PATENTS)
41
+
42
+ </div>
43
 
44
  ---
45
 
46
+ ## crdt-merge
47
+
48
+ Every standard merge strategy fails at least one of the three algebraic laws required for distributed convergence. That failure is mathematical, not implementational.
49
+
50
+ **crdt-merge** is the fix. A patented architecture that makes *any* merge operation -- across model weights, datasets, or agent memory -- fully conflict-free, order-independent, and provably convergent.
51
 
52
+ 26 strategies. Zero coordination. One correct result.
53
 
54
+ ```
55
+ pip install crdt-merge
56
+ ```
57
 
58
+ **[GitHub](https://github.com/mgillr/crdt-merge)** · **[PyPI](https://pypi.org/project/crdt-merge/)** · **[Architecture](https://github.com/mgillr/crdt-merge/blob/main/docs/CRDT_ARCHITECTURE.md)** · **[Download Stats](https://clickpy.clickhouse.com/dashboard/crdt-merge)**
59
 
60
  ---
61
 
62
+ ## Research Directions
63
+
64
+ We are working on problems at the intersection of distributed systems theory and machine intelligence:
65
+
66
+ - **Convergent model merging** -- deterministic, algebraically verified, at arbitrary scale
67
+ - **Distributed trust and verification** -- Byzantine-resilient protocols for open networks
68
+ - **Provenance-complete AI** -- cryptographic audit trails for every merge decision
69
+ - **Continual learning** -- knowledge acquisition without catastrophic forgetting
70
+ - **Multi-agent coordination** -- convergent shared state without central arbiters
71
 
72
+ Selected findings are published alongside the library. More to follow.
 
 
 
 
 
 
 
73
 
74
  ---
75
 
76
+ ## Live Demonstrations
77
 
78
+ | | |
79
+ |:--|:--|
80
+ | **[crdt-merge](https://huggingface.co/spaces/Optitransfer/crdt-merge)** | Merge real models with full provenance |
81
+ | **[Convergence Lab](https://huggingface.co/spaces/Optitransfer/convergence-lab)** | 26 strategies, convergence visualisation, live experiments |
82
+ | **[Data Playground](https://huggingface.co/spaces/Optitransfer/crdt-merge-data)** | CRDT merging on tabular data |
83
+ | **[Federation](https://huggingface.co/spaces/Optitransfer/crdt-merge-federation)** | Multi-node gossip convergence |
84
 
85
  ---
86
 
87
+ ## By the Numbers
88
 
89
+ | | |
90
+ |:--|--:|
91
+ | Merge strategies (all CRDT-verified) | **26** |
92
+ | Test suite | **4,498 passing** |
93
+ | CRDT overhead per merge | **< 0.5 ms** |
94
+ | Acceleration vs naive merge | **38.8×** |
95
+ | Cross-language ports | **Python · Rust · TypeScript · Java** |
96
+ | Encryption backends | **4** |
97
 
98
+ ---
 
99
 
100
+ ## Links
101
+
102
+ | | |
103
+ |:--|:--|
104
+ | **Source** | [github.com/mgillr/crdt-merge](https://github.com/mgillr/crdt-merge) |
105
+ | **Package** | [pypi.org/project/crdt-merge](https://pypi.org/project/crdt-merge/) |
106
+ | **Download Analytics** | [ClickPy](https://clickpy.clickhouse.com/dashboard/crdt-merge) · [PePy](https://pepy.tech/projects/crdt-merge) |
107
+ | **Rust Port** | [crates.io/crates/crdt-merge](https://crates.io/crates/crdt-merge) |
108
+ | **TypeScript Port** | [npmjs.com/package/crdt-merge](https://www.npmjs.com/package/crdt-merge) |
109
+ | **License** | BSL 1.1 → Apache 2.0 (March 2028) |
110
+ | **Patent** | UK Application GB 2607132.4 |
111
 
112
  ---
113
 
114
  ## Data Products
115
 
116
+ Optitransfer maintains compliance-ready, quality-scored web corpora for AI training.
117
 
118
+ | Dataset | Access |
119
+ |:--|:--|
120
+ | Swiss Web Premium (A+) -- 110K records, EU AI Act compliant | [Sample](https://huggingface.co/datasets/OptiTransferData/swiss-web-premium-ch) · [Full](https://huggingface.co/datasets/OptiTransferData/swiss-web-premium-ch-full) |
 
 
121
 
122
  ---
123
 
124
+ <div align="center">
125
+
126
+ **Switzerland**
127
 
128
+ [data@optitransfer.ch](mailto:data@optitransfer.ch)
 
 
 
129
 
130
+ </div>