Instructions to use Polydim-Einsof/Qwen2.5-Coder-32B-Einsof-LoRA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Polydim-Einsof/Qwen2.5-Coder-32B-Einsof-LoRA with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("unsloth/Qwen2.5-Coder-32B-Instruct-bnb-4bit") model = PeftModel.from_pretrained(base_model, "Polydim-Einsof/Qwen2.5-Coder-32B-Einsof-LoRA") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Unsloth Desktop
feat(indexing): sync CITATION.cff, codemeta.json & DOIs (10.5281/zenodo.21209973, hal-05847496v1)
Browse files- CITATION.cff +87 -0
- README.md +26 -0
- codemeta.json +105 -0
CITATION.cff
ADDED
|
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
cff-version: 1.2.0
|
| 2 |
+
message: "If you use POLYDIM, EinsofOS, or the PMTP Zero-Copy Stiefel Manifold IPC Architecture in your research or software, please cite it using the following metadata."
|
| 3 |
+
title: "Dimension Is All You Need: Eradicating the 1D Worm Collapse via Native High-Dimensional Manifold Inter-Process Communication in S^{D-1}"
|
| 4 |
+
type: software
|
| 5 |
+
version: "1550.26505"
|
| 6 |
+
date-released: "2026-11-01"
|
| 7 |
+
doi: "10.5281/zenodo.21136742"
|
| 8 |
+
url: "https://github.com/AGT1973/POLYDIM_EINSOF"
|
| 9 |
+
repository-code: "https://github.com/AGT1973/POLYDIM_EINSOF"
|
| 10 |
+
repository-artifact: "https://huggingface.co/Polydim-Einsof"
|
| 11 |
+
license: "MIT"
|
| 12 |
+
authors:
|
| 13 |
+
- family-names: "García Traba"
|
| 14 |
+
given-names: "Ariel H."
|
| 15 |
+
orcid: "https://orcid.org/0009-0001-2787-6067"
|
| 16 |
+
affiliation: "EinsofOS Independent Research / POLYDIM Research Initiative"
|
| 17 |
+
email: "polydim.einsof@gmail.com"
|
| 18 |
+
identifiers:
|
| 19 |
+
- type: doi
|
| 20 |
+
value: "10.5281/zenodo.21136742"
|
| 21 |
+
description: "CERN / Zenodo Canonical Preprint & Software Archive (Primary DOI)"
|
| 22 |
+
- type: doi
|
| 23 |
+
value: "10.5281/zenodo.21209973"
|
| 24 |
+
description: "CERN / Zenodo Extended Empirical & Theorem Verification Archive (Secondary DOI)"
|
| 25 |
+
- type: other
|
| 26 |
+
value: "hal-05847496v1"
|
| 27 |
+
description: "CNRS / HAL Open Science Archive (France)"
|
| 28 |
+
- type: url
|
| 29 |
+
value: "https://osf.io/za3s7"
|
| 30 |
+
description: "Center for Open Science / OSF Official Repository"
|
| 31 |
+
- type: other
|
| 32 |
+
value: "jxiv:7040"
|
| 33 |
+
description: "JST / Jxiv Preprint Server (Japan, Submission ID 7040)"
|
| 34 |
+
- type: url
|
| 35 |
+
value: "https://huggingface.co/Polydim-Einsof"
|
| 36 |
+
description: "Hugging Face Hub Official Organization & Stiefel Manifold LoRA Adapters"
|
| 37 |
+
keywords:
|
| 38 |
+
- "Riemannian Manifolds"
|
| 39 |
+
- "Stiefel Manifold"
|
| 40 |
+
- "High-Dimensional Computing"
|
| 41 |
+
- "Inter-Process Communication"
|
| 42 |
+
- "Shared Memory"
|
| 43 |
+
- "Information Theory"
|
| 44 |
+
- "Data Processing Inequality"
|
| 45 |
+
- "Multi-Agent Systems"
|
| 46 |
+
- "Zero-Copy IPC"
|
| 47 |
+
- "Clifford Algebra"
|
| 48 |
+
- "Vector AI"
|
| 49 |
+
- "POLYDIM"
|
| 50 |
+
- "EinsofOS"
|
| 51 |
+
- "PMTP"
|
| 52 |
+
abstract: >-
|
| 53 |
+
Contemporary Artificial Intelligence architectures universally serialize dense high-dimensional
|
| 54 |
+
neural representations (h in R^D, D >= 10^4) into flat 1D token sequences. Under Shannon's Data
|
| 55 |
+
Processing Inequality (DPI), this 1D Worm Paradigm inflicts irreversible information collapse,
|
| 56 |
+
injects 140 ms quadratic decoding latency per hop, and incurs massive financial costs. We introduce
|
| 57 |
+
POLYDIM / EinsofOS: a native high-dimensional computing paradigm where autonomous agents communicate
|
| 58 |
+
losslessly on continuous Riemannian manifolds (S^{D-1} and St(D, K)) via Zero-Copy Shared Memory
|
| 59 |
+
Inter-Process Communication (PMTP V1550+ and PMTP-v4 Triadic Fusion). Across 1,731 formal theorems
|
| 60 |
+
certified in PostgreSQL 16 and 26,505+ continuous physical cycles on host silicon, POLYDIM certifies
|
| 61 |
+
a data-path control latency of 34.8 microseconds (229.885 GB/s nominal payload-equivalent throughput
|
| 62 |
+
on host RAM, 139.78 GB/s sustained VRAM throughput on Dual Tesla T4 / Triton SRAM tiles), achieving
|
| 63 |
+
a 4,023x speedup over text tokenization with $0.00 marginal token cost and Neumaier compensated
|
| 64 |
+
orthonormality drift <= 10^{-15}.
|
| 65 |
+
preferred-citation:
|
| 66 |
+
type: article
|
| 67 |
+
title: "Dimension Is All You Need: Eradicating the 1D Worm Collapse via Native High-Dimensional Manifold Inter-Process Communication in S^{D-1}"
|
| 68 |
+
authors:
|
| 69 |
+
- family-names: "García Traba"
|
| 70 |
+
given-names: "Ariel H."
|
| 71 |
+
orcid: "https://orcid.org/0009-0001-2787-6067"
|
| 72 |
+
affiliation: "EinsofOS Independent Research / POLYDIM Research Initiative"
|
| 73 |
+
year: 2026
|
| 74 |
+
month: 11
|
| 75 |
+
doi: "10.5281/zenodo.21136742"
|
| 76 |
+
url: "https://doi.org/10.5281/zenodo.21136742"
|
| 77 |
+
identifiers:
|
| 78 |
+
- type: doi
|
| 79 |
+
value: "10.5281/zenodo.21136742"
|
| 80 |
+
- type: doi
|
| 81 |
+
value: "10.5281/zenodo.21209973"
|
| 82 |
+
- type: other
|
| 83 |
+
value: "hal-05847496v1"
|
| 84 |
+
- type: url
|
| 85 |
+
value: "https://osf.io/za3s7"
|
| 86 |
+
- type: other
|
| 87 |
+
value: "jxiv:7040"
|
README.md
CHANGED
|
@@ -42,3 +42,29 @@ model, tokenizer = FastLanguageModel.from_pretrained(
|
|
| 42 |
)
|
| 43 |
FastLanguageModel.for_inference(model)
|
| 44 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
)
|
| 43 |
FastLanguageModel.for_inference(model)
|
| 44 |
```
|
| 45 |
+
|
| 46 |
+
---
|
| 47 |
+
|
| 48 |
+
## Global Scientific Indexing & Persistent Identifiers (CERN Zenodo / CNRS HAL / JST Jxiv / OSF / ORCID)
|
| 49 |
+
|
| 50 |
+
- **Author:** Ariel H. García Traba ([ORCID: 0009-0001-2787-6067](https://orcid.org/0009-0001-2787-6067))
|
| 51 |
+
- **Affiliation:** EinsofOS Independent Research / POLYDIM Research Initiative
|
| 52 |
+
- **CERN / Zenodo Primary DOI:** [10.5281/zenodo.21136742](https://doi.org/10.5281/zenodo.21136742)
|
| 53 |
+
- **CERN / Zenodo Secondary DOI:** [10.5281/zenodo.21209973](https://doi.org/10.5281/zenodo.21209973)
|
| 54 |
+
- **CNRS / HAL Open Science (France):** [`hal-05847496v1`](https://hal.science/hal-05847496v1)
|
| 55 |
+
- **Center for Open Science (OSF):** [`https://osf.io/za3s7`](https://osf.io/za3s7)
|
| 56 |
+
- **JST / Jxiv Preprint Server (Japan):** `Jxiv ID 7040`
|
| 57 |
+
- **Canonical Papers & LaTeX Archive:** [`Polydim-Einsof/POLYDIM-EINSOF-Canonical-Papers-V1550`](https://huggingface.co/datasets/Polydim-Einsof/POLYDIM-EINSOF-Canonical-Papers-V1550)
|
| 58 |
+
|
| 59 |
+
```bibtex
|
| 60 |
+
@article{garciatraba2026polydim,
|
| 61 |
+
title = {Dimension Is All You Need: Eradicating the 1D Worm Collapse via Native High-Dimensional Manifold Inter-Process Communication in S^{D-1}},
|
| 62 |
+
author = {Garc{\'i}a Traba, Ariel H.},
|
| 63 |
+
year = {2026},
|
| 64 |
+
month = {nov},
|
| 65 |
+
publisher = {Zenodo / CERN & HAL Open Science},
|
| 66 |
+
doi = {10.5281/zenodo.21209973},
|
| 67 |
+
url = {https://doi.org/10.5281/zenodo.21209973},
|
| 68 |
+
note = {Primary DOI: 10.5281/zenodo.21136742; HAL ID: hal-05847496v1; OSF: https://osf.io/za3s7; Jxiv ID: 7040; ORCID: 0009-0001-2787-6067; Affiliation: EinsofOS Independent Research / POLYDIM Research Initiative}
|
| 69 |
+
}
|
| 70 |
+
```
|
codemeta.json
ADDED
|
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"@context": [
|
| 3 |
+
"https://doi.org/10.5063/schema/codemeta-2.0",
|
| 4 |
+
"https://schema.org"
|
| 5 |
+
],
|
| 6 |
+
"@type": [
|
| 7 |
+
"SoftwareSourceCode",
|
| 8 |
+
"ScholarlyArticle"
|
| 9 |
+
],
|
| 10 |
+
"@id": "https://doi.org/10.5281/zenodo.21136742",
|
| 11 |
+
"identifier": [
|
| 12 |
+
{
|
| 13 |
+
"@type": "PropertyValue",
|
| 14 |
+
"propertyID": "DOI",
|
| 15 |
+
"value": "10.5281/zenodo.21136742",
|
| 16 |
+
"url": "https://doi.org/10.5281/zenodo.21136742"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"@type": "PropertyValue",
|
| 20 |
+
"propertyID": "DOI",
|
| 21 |
+
"value": "10.5281/zenodo.21209973",
|
| 22 |
+
"url": "https://doi.org/10.5281/zenodo.21209973"
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"@type": "PropertyValue",
|
| 26 |
+
"propertyID": "HAL",
|
| 27 |
+
"value": "hal-05847496v1",
|
| 28 |
+
"url": "https://hal.science/hal-05847496v1"
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"@type": "PropertyValue",
|
| 32 |
+
"propertyID": "OSF",
|
| 33 |
+
"value": "za3s7",
|
| 34 |
+
"url": "https://osf.io/za3s7"
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"@type": "PropertyValue",
|
| 38 |
+
"propertyID": "Jxiv",
|
| 39 |
+
"value": "7040",
|
| 40 |
+
"url": "https://jxiv.jst.go.jp"
|
| 41 |
+
}
|
| 42 |
+
],
|
| 43 |
+
"name": "POLYDIM / EinsofOS: Native High-Dimensional Stiefel Manifold IPC & Vector AI Architecture",
|
| 44 |
+
"headline": "Dimension Is All You Need: Eradicating the 1D Worm Collapse via Native High-Dimensional Manifold Inter-Process Communication in S^{D-1}",
|
| 45 |
+
"version": "1550.26505",
|
| 46 |
+
"datePublished": "2026-11-01",
|
| 47 |
+
"dateCreated": "2026-10-01",
|
| 48 |
+
"dateModified": "2026-11-01",
|
| 49 |
+
"license": "https://spdx.org/licenses/MIT",
|
| 50 |
+
"codeRepository": "https://github.com/AGT1973/POLYDIM_EINSOF",
|
| 51 |
+
"url": "https://huggingface.co/Polydim-Einsof",
|
| 52 |
+
"sameAs": [
|
| 53 |
+
"https://doi.org/10.5281/zenodo.21136742",
|
| 54 |
+
"https://doi.org/10.5281/zenodo.21209973",
|
| 55 |
+
"https://hal.science/hal-05847496v1",
|
| 56 |
+
"https://osf.io/za3s7",
|
| 57 |
+
"https://orcid.org/0009-0001-2787-6067",
|
| 58 |
+
"https://huggingface.co/Polydim-Einsof",
|
| 59 |
+
"https://github.com/AGT1973/POLYDIM_EINSOF"
|
| 60 |
+
],
|
| 61 |
+
"author": [
|
| 62 |
+
{
|
| 63 |
+
"@type": "Person",
|
| 64 |
+
"@id": "https://orcid.org/0009-0001-2787-6067",
|
| 65 |
+
"givenName": "Ariel H.",
|
| 66 |
+
"familyName": "García Traba",
|
| 67 |
+
"name": "Ariel H. García Traba",
|
| 68 |
+
"email": "polydim.einsof@gmail.com",
|
| 69 |
+
"affiliation": {
|
| 70 |
+
"@type": "Organization",
|
| 71 |
+
"name": "EinsofOS Independent Research / POLYDIM Research Initiative"
|
| 72 |
+
}
|
| 73 |
+
}
|
| 74 |
+
],
|
| 75 |
+
"programmingLanguage": [
|
| 76 |
+
"C++20",
|
| 77 |
+
"Rust 2021",
|
| 78 |
+
"CUDA",
|
| 79 |
+
"Triton",
|
| 80 |
+
"Python 3",
|
| 81 |
+
"Dart 3",
|
| 82 |
+
"Lean 4"
|
| 83 |
+
],
|
| 84 |
+
"operatingSystem": [
|
| 85 |
+
"Linux (Debian 13)",
|
| 86 |
+
"Windows 11"
|
| 87 |
+
],
|
| 88 |
+
"keywords": [
|
| 89 |
+
"Riemannian Manifolds",
|
| 90 |
+
"Stiefel Manifold",
|
| 91 |
+
"High-Dimensional Computing",
|
| 92 |
+
"Inter-Process Communication",
|
| 93 |
+
"Shared Memory",
|
| 94 |
+
"Information Theory",
|
| 95 |
+
"Data Processing Inequality",
|
| 96 |
+
"Multi-Agent Systems",
|
| 97 |
+
"Zero-Copy IPC",
|
| 98 |
+
"Clifford Algebra",
|
| 99 |
+
"Vector AI",
|
| 100 |
+
"POLYDIM",
|
| 101 |
+
"EinsofOS",
|
| 102 |
+
"PMTP"
|
| 103 |
+
],
|
| 104 |
+
"description": "Contemporary Artificial Intelligence architectures universally serialize dense high-dimensional neural representations (h in R^D, D >= 10^4) into flat 1D token sequences. Under Shannon's Data Processing Inequality (DPI), this 1D Worm Paradigm inflicts irreversible information collapse, injects 140 ms quadratic decoding latency per hop, and incurs massive financial costs. POLYDIM / EinsofOS introduces a native high-dimensional computing paradigm where autonomous agents communicate losslessly on continuous Riemannian manifolds (S^{D-1} and St(D, K)) via Zero-Copy Shared Memory Inter-Process Communication (PMTP V1550+ and PMTP-v4 Triadic Fusion), achieving 34.8 microseconds control latency, 229.885 GB/s nominal payload-equivalent throughput on host RAM, 139.78 GB/s sustained VRAM throughput on Dual Tesla T4 / Triton SRAM tiles, and Neumaier compensated orthonormality drift <= 10^{-15}."
|
| 105 |
+
}
|