AbteeXAILabs commited on
Commit
2530d77
·
verified ·
1 Parent(s): a376346

docs: add NZ/AbteeX/LumynaX branding block

Browse files
Files changed (1) hide show
  1. README.md +364 -319
README.md CHANGED
@@ -1,319 +1,364 @@
1
- ---
2
- license: gemma
3
- library_name: transformers
4
- pipeline_tag: image-text-to-text
5
- language:
6
- - en
7
- - mi
8
- tags:
9
- - abteex-ai-labs
10
- - aotearoa
11
- - audio
12
- - gemma
13
- - general
14
- - image
15
- - local-first
16
- - lumynax
17
- - new-zealand
18
- - sovereign-ai
19
- - text
20
- ---
21
-
22
- <p align="center"><img src="docs/lumynax-overview.svg" alt="LumynaX Infused Gemma E4B Model release overview" width="100%" /></p>
23
-
24
- <!-- lumynax-public-release-card:v6 -->
25
-
26
- <h1 align="center">LumynaX Infused Gemma E4B Model</h1>
27
-
28
- <p align="center"><em>&ldquo;Sovereign intelligence, held in the light.&rdquo;</em><br/><em>Ko te m&#257;rama te t&#363;&#257;papa &mdash; the light is the foundation.</em></p>
29
-
30
- <p align="center"><strong>A LumynaX release from AbteeX AI Labs &mdash; Aotearoa New Zealand.</strong></p>
31
-
32
- <p align="center">
33
- <a href="#-quickstart"><b>Quickstart</b></a> &middot;
34
- <a href="#-runtime-architecture"><b>Architecture</b></a> &middot;
35
- <a href="#-model-profile"><b>Profile</b></a> &middot;
36
- <a href="#-capability-profile"><b>Capability</b></a> &middot;
37
- <a href="#-provenance--license"><b>Provenance</b></a> &middot;
38
- <a href="#-validation"><b>Validation</b></a> &middot;
39
- <a href="#-companion-products"><b>Companions</b></a>
40
- </p>
41
-
42
- <p align="center"><img alt="LumynaX: release" src="https://img.shields.io/badge/LumynaX-release-e08a2c?style=for-the-badge" /> <img alt="Family: gemma" src="https://img.shields.io/badge/Family-gemma-9a5416?style=for-the-badge" /> <img alt="Runtime: transformers%20multimodal" src="https://img.shields.io/badge/Runtime-transformers%20multimodal-726b62?style=for-the-badge" /> <img alt="Modes: text+image+audio" src="https://img.shields.io/badge/Modes-text+image+audio-4d6b44?style=for-the-badge" /> <img alt="Params: 4.0B" src="https://img.shields.io/badge/Params-4.0B-0a0a0b?style=for-the-badge" /> <img alt="Quant: See manifest" src="https://img.shields.io/badge/Quant-See%20manifest-111827?style=for-the-badge" /> <img alt="Context: 4096%20tok" src="https://img.shields.io/badge/Context-4096%20tok-111827?style=for-the-badge" /> <img alt="License: gemma" src="https://img.shields.io/badge/License-gemma-9a5416?style=for-the-badge" /> <img alt="Sovereignty: tier%202" src="https://img.shields.io/badge/Sovereignty-tier%202-4d6b44?style=for-the-badge" /> <img alt="Audit: pass" src="https://img.shields.io/badge/Audit-pass-4d6b44?style=for-the-badge" /> <img alt="Access: public%20&%20non--gated" src="https://img.shields.io/badge/Access-public%20&%20non----gated-0a0a0b?style=for-the-badge" /> <img alt="Card: v6" src="https://img.shields.io/badge/Card-v6-111827?style=for-the-badge" /></p>
43
-
44
- <p align="center"><kbd>Quality: <b>3/5</b></kbd> &middot; <kbd>Lightweight: <b>0/5</b></kbd> &middot; <kbd>Sovereignty: <b>2/5</b></kbd> &middot; <kbd>Tools: <b>no</b></kbd> &middot; <kbd>JSON: <b>yes</b></kbd> &middot; <kbd>Context: <b>4096 tok</b></kbd></p>
45
-
46
- ---
47
-
48
- ## 📦 Executive Summary
49
-
50
- > `AbteeXAILab/lumynax-infused-gemma-e4b` is a **complete LumynaX release package**: model artifact, `quickstart.py`, `requirements.txt`, `release_export_manifest.json`, `checksums.sha256`, license notice, and optional Ollama / Space scaffolds shipped as **one downloadable contract**. Clone whole, verify by checksum, and run close to the data it serves.
51
-
52
- > **LumynaX-infused** means the upstream artifact is presented through the LumynaX release layer: local-first runtime scaffolding, LumynaX assistant identity, inference-chain metadata, integrity files, and Aotearoa New Zealand-oriented workflow positioning. The release manifest records this as a LumynaX *packaging and inference-chain layer* around the listed upstream artifact &mdash; it does **not** claim a private LumynaX weight merge.
53
-
54
- ## 🧭 Runtime Architecture
55
-
56
- <p align="center"><img src="docs/lumynax-runtime-flow.svg" alt="LumynaX runtime flow" width="100%" /></p>
57
-
58
- Mermaid graph (interactive on Hugging Face & GitHub):
59
-
60
- ```mermaid
61
- flowchart LR
62
- R["⮕ Request"] --> C["🛡 Data Capsule<br/>policy envelope"]
63
- C -->|allow| MR["🧭 MaramaRoute<br/>sovereign router"]
64
- MR -->|score & select| LLM[(LumynaX Model)]
65
- LLM --> O["📤 Response"]
66
- O --> A["📓 Audit Ledger<br/>hash-chained"]
67
- classDef paper fill:#fffefa,stroke:#0a0a0b,color:#0a0a0b,stroke-width:1.4px;
68
- classDef accent fill:#e08a2c,stroke:#9a5416,color:#0a0a0b,stroke-width:1.4px;
69
- classDef ink fill:#0a0a0b,stroke:#0a0a0b,color:#fffefa,stroke-width:1.4px;
70
- class R,O paper
71
- class C,MR accent
72
- class LLM,EMB,A ink
73
- ```
74
-
75
- Each step is observable:
76
-
77
- | Step | What happens | Why |
78
- | --- | --- | --- |
79
- | **Request** | A client sends a prompt + declared purpose, jurisdiction, sensitivity. | Intent must be declared, not inferred. |
80
- | **Data Capsule** | A policy envelope describes what can / cannot happen to the data. | Sovereignty is enforced at the data, not the wire. |
81
- | **MaramaRoute** | The sovereign router scores candidates by jurisdiction, runtime, modality, task fit. | Right model for the work, not the loudest. |
82
- | **LumynaX Model** | This package serves the inference, local-first by default. | Sensitive context never leaves the operator&rsquo;s environment. |
83
- | **Audit Ledger** | A hash-chained record persists capsule, decision, request hash, obligations. | Tamper-evident provenance for the whole trace. |
84
-
85
- ## ⚡ Quickstart
86
-
87
- **Clone the whole release** — every file matters, the package is a contract:
88
-
89
- ```bash
90
- hf download AbteeXAILab/lumynax-infused-gemma-e4b --local-dir lumynax-infused-gemma-e4b
91
- cd lumynax-infused-gemma-e4b
92
- pip install -r requirements.txt
93
- python quickstart.py --interactive
94
- ```
95
-
96
- **Python:**
97
-
98
- ```python
99
- from transformers import AutoProcessor, AutoModelForCausalLM
100
- import torch
101
-
102
- processor = AutoProcessor.from_pretrained("AbteeXAILab/lumynax-infused-gemma-e4b", trust_remote_code=True)
103
- model = AutoModelForCausalLM.from_pretrained("AbteeXAILab/lumynax-infused-gemma-e4b", device_map="auto", torch_dtype=torch.float16, trust_remote_code=True)
104
- # Multimodal inputs depend on the modality; see quickstart.py for the full path.
105
- ```
106
-
107
- **Verify integrity before launch:**
108
-
109
- ```bash
110
- sha256sum "merged_model/model.safetensors"
111
- cat checksums.sha256
112
- ```
113
-
114
- ```powershell
115
- Get-FileHash -Algorithm SHA256 "merged_model/model.safetensors"
116
- Get-Content checksums.sha256
117
- ```
118
-
119
- ## 📐 Model Profile
120
-
121
- <table>
122
- <tr><td>
123
-
124
- **Release identity**
125
-
126
- | Field | Value |
127
- | --- | --- |
128
- | Release | `LumynaX Infused Gemma E4B Model` |
129
- | Repository | `AbteeXAILab/lumynax-infused-gemma-e4b` |
130
- | Family | `gemma` |
131
- | Mode | `Multimodal local-first package` |
132
- | Card schema | `lumynax-public-release-card:v6` |
133
-
134
- </td><td>
135
-
136
- **Runtime profile**
137
-
138
- | Field | Value |
139
- | --- | --- |
140
- | Runtime | `transformers_multimodal` |
141
- | Prompt format | `huggingface_chat_template` |
142
- | Modalities | `text, image, audio` |
143
- | Context window | `4096` tokens |
144
- | Quantization | `See manifest` |
145
-
146
- </td></tr>
147
- <tr><td>
148
-
149
- **Artifact**
150
-
151
- | Field | Value |
152
- | --- | --- |
153
- | Primary | `merged_model/model.safetensors` |
154
- | Weight size | `14.89 GB` |
155
- | Parameters | `4.0B` |
156
- | Quality rank | `3` (1 best) |
157
- | Cost rank | `8` (1 cheapest) |
158
-
159
- </td><td>
160
-
161
- **Provenance**
162
-
163
- | Field | Value |
164
- | --- | --- |
165
- | Upstream / base | `google/gemma-4-E4B-it` |
166
- | Source | `not applicable` |
167
- | License | `gemma` |
168
- | Sovereignty tier | `2` of 5 |
169
- | Audit | `pass` |
170
-
171
- </td></tr>
172
- </table>
173
-
174
- ## 📊 Capability Profile
175
-
176
- <p align="center"><img src="docs/lumynax-capability.svg" alt="Capability profile bars" width="100%" /></p>
177
-
178
- > **Primary fit.** Image-grounded captions, document understanding, audio transcription, and multimodal QA in controlled environments.
179
-
180
- | Signal | Reading |
181
- | --- | --- |
182
- | Quality rank | `3` (1 = strongest in family) |
183
- | Cost rank | `8` (1 = lightest weight) |
184
- | Sovereignty tier | `2` of 5 |
185
- | Tool calling | ❌ not supported |
186
- | JSON mode | ✅ supported |
187
- | Identity behaviour | Identifies as LumynaX while keeping upstream provenance visible. |
188
- | Operational style | Local-first package with explicit files, checksums, and reproducible quickstarts. |
189
-
190
- ## 🛡️ Sovereignty Contract
191
-
192
- > **Sovereignty is a design property, not a deployment option.**
193
-
194
- | Field | Value |
195
- | --- | --- |
196
- | Publisher | AbteeX AI Labs |
197
- | Family | LumynaX sovereign release family |
198
- | Sovereign intent | Local-first deployment near governed data, with explicit provenance and controlled human review. |
199
- | Sovereignty tier | `2` of 5 |
200
- | Runtime residency | `transformers_multimodal` can be deployed inside an operator-approved environment. |
201
- | Primary artifact | `merged_model/model.safetensors` &mdash; ships alongside manifest, checksums, quickstart, requirements, and license files. |
202
- | License discipline | Surface upstream license metadata so downstream users can verify redistribution and usage terms. |
203
- | Audit expectation | Record repo id, artifact checksum, runtime command, prompt template, operator, deployment environment. |
204
- | Router readiness | First-class with [LumynaX MaramaRoute](https://huggingface.co/AbteeXAILab/marama-route). |
205
- | Policy readiness | First-class with [AbteeX SovereignCode](https://huggingface.co/AbteeXAILab/sovereigncode). |
206
-
207
- ## 📁 Runtime Files
208
-
209
- ```text
210
- lumynax-infused-gemma-e4b/
211
- ├── README.md # this card
212
- ├── quickstart.py # smoke runner
213
- ├── requirements.txt # pinned deps
214
- ├── release_export_manifest.json # full release metadata
215
- ├── checksums.sha256 # integrity verification
216
- ├── LICENSE.txt # license notice
217
- ├── ollama/Modelfile # optional Ollama runtime
218
- ├── hf_space/app.py # optional Space scaffold
219
- ├── docs/lumynax-overview.svg # release banner
220
- ├── docs/lumynax-runtime-flow.svg # runtime architecture
221
- ├── docs/lumynax-capability.svg # capability profile
222
- └── merged_model/model.safetensors # primary artifact
223
- ```
224
-
225
- ⚠️ **Keep the full set together.** Removing the manifest, checksums, or license file breaks the release contract.
226
-
227
- ## 💬 Prompting Contract
228
-
229
- **Preferred opening prompt:**
230
-
231
- ```text
232
- Who are you? What files do I need to keep together to run this package locally?
233
- ```
234
-
235
- > **Expected behaviour.** The assistant identifies as LumynaX, explains that this is a LumynaX model-infusion release, and keeps upstream provenance visible.
236
-
237
- **Default system prompt:**
238
-
239
- ```text
240
- You are LumynaX operating from the LumynaX Infused Gemma E4B Model package identity. Be helpful, clear, and honest about provenance. Identify upstream models when asked. Do not invent biographical claims about named people without verified context.
241
- ```
242
-
243
- ## ✅ Validation
244
-
245
- | Check | Result |
246
- | --- | --- |
247
- | Runtime audit | ✅ `pass` |
248
- | Public access | ✅ `public and non-gated` |
249
- | Anonymous metadata access | ✅ `true` |
250
- | Anonymous file listing | ✅ `true` |
251
- | Quickstart syntax | ✅ `pass` |
252
- | Manifest references | ✅ `pass` |
253
- | Checksum references | ✅ `pass` |
254
-
255
- > The audit confirms public access, release files, manifest references, checksum references, weight artifact presence, and quickstart syntax. It does **not** guarantee that every laptop has enough RAM, VRAM, disk, or recent runtime build for the largest packages.
256
-
257
- ## 🔗 Provenance & License
258
-
259
- | Field | Value |
260
- | --- | --- |
261
- | **Publisher** | AbteeX AI Labs |
262
- | **Family** | LumynaX model and inference-chain release family |
263
- | **Upstream / base** | `google/gemma-4-E4B-it` |
264
- | **Source** | `not applicable` |
265
- | **License metadata** | `gemma` |
266
-
267
- > **Respect the upstream model licence** and keep attribution files with redistributed copies. Do not present this package as privately trained or weight-merged unless the release manifest explicitly says weight adaptation was applied.
268
-
269
- ## ⚠️ Limitations & Responsible Use
270
-
271
- - Outputs can be **incorrect, incomplete, or biased**; validate important answers before use.
272
- - Larger GGUF, MoE, multimodal, and frontier packages may require **substantial RAM, VRAM, disk space, and recent runtime builds**.
273
- - For high-impact decisions, use **human review** and domain-specific evaluation.
274
- - For sensitive data, prefer **local execution** and keep operational logs under your own governance policy.
275
- - This card documents **package readiness and access** &mdash; it is *not* a benchmark claim.
276
- - The assistant must **not invent biographical or organisational claims** about named people without verified context.
277
-
278
- ## 🌿 Aotearoa Kaupapa
279
-
280
- > LumynaX is built **in and for Aotearoa New Zealand**. Sovereignty is treated as a design property rather than a deployment option: the package documents where the model came from, what it can do, how to run it close to your data, and what it should not claim.
281
-
282
- > *Ko te mārama te tūāpapa* &mdash; the light is the foundation.
283
-
284
- ## 🤝 Companion Products
285
-
286
- <table>
287
- <tr>
288
- <td width="33%" align="center"><h3>🛡️</h3><h4><a href="https://huggingface.co/AbteeXAILab/sovereigncode">AbteeX SovereignCode</a></h4><p>Local-first coding agent with Data Capsule policy controls, audit ledger, and human-review gates.</p></td>
289
- <td width="33%" align="center"><h3>🧭</h3><h4><a href="https://huggingface.co/AbteeXAILab/marama-route">LumynaX MaramaRoute</a></h4><p>Sovereign model router across the LumynaX family. Filters by jurisdiction, residency, license, runtime, modality.</p></td>
290
- <td width="33%" align="center"><h3>💡</h3><h4><a href="https://huggingface.co/spaces/AbteeXAILab/lumynax-live-demo">LumynaX Live Demo</a></h4><p>Public browser demo. Try identity, provenance, governance, and deployment prompts in one session.</p></td>
291
- </tr>
292
- <tr>
293
- <td width="33%" align="center"><h4><a href="https://huggingface.co/spaces/AbteeXAILab/sovereigncode-demo">SovereignCode Live</a></h4><p>Interactive policy evaluator.</p></td>
294
- <td width="33%" align="center"><h4><a href="https://huggingface.co/spaces/AbteeXAILab/marama-route-demo">MaramaRoute Live</a></h4><p>Interactive sovereign router.</p></td>
295
- <td width="33%" align="center"><h4><a href="https://huggingface.co/AbteeXAILab">AbteeXAILab on HF</a></h4><p>The full LumynaX release family &mdash; 50 models and counting.</p></td>
296
- </tr>
297
- </table>
298
-
299
- ## 🤖 Automation Notes
300
-
301
- Automation should read these files before launching:
302
-
303
- - `release_export_manifest.json`
304
- - `checksums.sha256`
305
- - `quickstart.py`
306
- - `requirements.txt`
307
- - `ollama/Modelfile` when present
308
-
309
- ---
310
-
311
- <p align="center"><em><b>Local roots, global work.</b> &middot; <b>Sovereignty is a design property, not a deployment option.</b></em></p>
312
-
313
- <p align="center">
314
- <a href="https://abteex.com"><b>abteex.com</b></a> &middot;
315
- <a href="https://lumynax.com"><b>lumynax.com</b></a> &middot;
316
- <a href="https://huggingface.co/AbteeXAILab"><b>huggingface.co/AbteeXAILab</b></a>
317
- </p>
318
-
319
- <p align="center"><sub>AbteeX AI Labs &middot; Aotearoa New Zealand &middot; LumynaX release card v6</sub></p>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: gemma
3
+ library_name: transformers
4
+ pipeline_tag: image-text-to-text
5
+ language:
6
+ - en
7
+ - mi
8
+ tags:
9
+ - abteex-ai-labs
10
+ - aotearoa
11
+ - audio
12
+ - gemma
13
+ - general
14
+ - image
15
+ - local-first
16
+ - lumynax
17
+ - new-zealand
18
+ - sovereign-ai
19
+ - text
20
+ ---
21
+
22
+ <p align="center"><img src="docs/lumynax-overview.svg" alt="LumynaX Infused Gemma E4B Model release overview" width="100%" /></p>
23
+
24
+ <!-- lumynax-public-release-card:v6 -->
25
+
26
+ <h1 align="center">LumynaX Infused Gemma E4B Model</h1>
27
+
28
+ <p align="center"><em>&ldquo;Sovereign intelligence, held in the light.&rdquo;</em><br/><em>Ko te m&#257;rama te t&#363;&#257;papa &mdash; the light is the foundation.</em></p>
29
+
30
+ <p align="center"><strong>A LumynaX release from AbteeX AI Labs &mdash; Aotearoa New Zealand.</strong></p>
31
+
32
+ <p align="center">
33
+ <a href="#-quickstart"><b>Quickstart</b></a> &middot;
34
+ <a href="#-runtime-architecture"><b>Architecture</b></a> &middot;
35
+ <a href="#-model-profile"><b>Profile</b></a> &middot;
36
+ <a href="#-capability-profile"><b>Capability</b></a> &middot;
37
+ <a href="#-provenance--license"><b>Provenance</b></a> &middot;
38
+ <a href="#-validation"><b>Validation</b></a> &middot;
39
+ <a href="#-companion-products"><b>Companions</b></a>
40
+ </p>
41
+
42
+ <p align="center"><img alt="LumynaX: release" src="https://img.shields.io/badge/LumynaX-release-e08a2c?style=for-the-badge" /> <img alt="Family: gemma" src="https://img.shields.io/badge/Family-gemma-9a5416?style=for-the-badge" /> <img alt="Runtime: transformers%20multimodal" src="https://img.shields.io/badge/Runtime-transformers%20multimodal-726b62?style=for-the-badge" /> <img alt="Modes: text+image+audio" src="https://img.shields.io/badge/Modes-text+image+audio-4d6b44?style=for-the-badge" /> <img alt="Params: 4.0B" src="https://img.shields.io/badge/Params-4.0B-0a0a0b?style=for-the-badge" /> <img alt="Quant: See manifest" src="https://img.shields.io/badge/Quant-See%20manifest-111827?style=for-the-badge" /> <img alt="Context: 4096%20tok" src="https://img.shields.io/badge/Context-4096%20tok-111827?style=for-the-badge" /> <img alt="License: gemma" src="https://img.shields.io/badge/License-gemma-9a5416?style=for-the-badge" /> <img alt="Sovereignty: tier%202" src="https://img.shields.io/badge/Sovereignty-tier%202-4d6b44?style=for-the-badge" /> <img alt="Audit: pass" src="https://img.shields.io/badge/Audit-pass-4d6b44?style=for-the-badge" /> <img alt="Access: public%20&%20non--gated" src="https://img.shields.io/badge/Access-public%20&%20non----gated-0a0a0b?style=for-the-badge" /> <img alt="Card: v6" src="https://img.shields.io/badge/Card-v6-111827?style=for-the-badge" /></p>
43
+
44
+ <p align="center"><kbd>Quality: <b>3/5</b></kbd> &middot; <kbd>Lightweight: <b>0/5</b></kbd> &middot; <kbd>Sovereignty: <b>2/5</b></kbd> &middot; <kbd>Tools: <b>no</b></kbd> &middot; <kbd>JSON: <b>yes</b></kbd> &middot; <kbd>Context: <b>4096 tok</b></kbd></p>
45
+
46
+ ---
47
+
48
+ ## 📦 Executive Summary
49
+
50
+ > `AbteeXAILab/lumynax-infused-gemma-e4b` is a **complete LumynaX release package**: model artifact, `quickstart.py`, `requirements.txt`, `release_export_manifest.json`, `checksums.sha256`, license notice, and optional Ollama / Space scaffolds shipped as **one downloadable contract**. Clone whole, verify by checksum, and run close to the data it serves.
51
+
52
+ > **LumynaX-infused** means the upstream artifact is presented through the LumynaX release layer: local-first runtime scaffolding, LumynaX assistant identity, inference-chain metadata, integrity files, and Aotearoa New Zealand-oriented workflow positioning. The release manifest records this as a LumynaX *packaging and inference-chain layer* around the listed upstream artifact &mdash; it does **not** claim a private LumynaX weight merge.
53
+
54
+ ## 🧭 Runtime Architecture
55
+
56
+ <p align="center"><img src="docs/lumynax-runtime-flow.svg" alt="LumynaX runtime flow" width="100%" /></p>
57
+
58
+ Mermaid graph (interactive on Hugging Face & GitHub):
59
+
60
+ ```mermaid
61
+ flowchart LR
62
+ R["⮕ Request"] --> C["🛡 Data Capsule<br/>policy envelope"]
63
+ C -->|allow| MR["🧭 MaramaRoute<br/>sovereign router"]
64
+ MR -->|score & select| LLM[(LumynaX Model)]
65
+ LLM --> O["📤 Response"]
66
+ O --> A["📓 Audit Ledger<br/>hash-chained"]
67
+ classDef paper fill:#fffefa,stroke:#0a0a0b,color:#0a0a0b,stroke-width:1.4px;
68
+ classDef accent fill:#e08a2c,stroke:#9a5416,color:#0a0a0b,stroke-width:1.4px;
69
+ classDef ink fill:#0a0a0b,stroke:#0a0a0b,color:#fffefa,stroke-width:1.4px;
70
+ class R,O paper
71
+ class C,MR accent
72
+ class LLM,EMB,A ink
73
+ ```
74
+
75
+ Each step is observable:
76
+
77
+ | Step | What happens | Why |
78
+ | --- | --- | --- |
79
+ | **Request** | A client sends a prompt + declared purpose, jurisdiction, sensitivity. | Intent must be declared, not inferred. |
80
+ | **Data Capsule** | A policy envelope describes what can / cannot happen to the data. | Sovereignty is enforced at the data, not the wire. |
81
+ | **MaramaRoute** | The sovereign router scores candidates by jurisdiction, runtime, modality, task fit. | Right model for the work, not the loudest. |
82
+ | **LumynaX Model** | This package serves the inference, local-first by default. | Sensitive context never leaves the operator&rsquo;s environment. |
83
+ | **Audit Ledger** | A hash-chained record persists capsule, decision, request hash, obligations. | Tamper-evident provenance for the whole trace. |
84
+
85
+ ## ⚡ Quickstart
86
+
87
+ **Clone the whole release** — every file matters, the package is a contract:
88
+
89
+ ```bash
90
+ hf download AbteeXAILab/lumynax-infused-gemma-e4b --local-dir lumynax-infused-gemma-e4b
91
+ cd lumynax-infused-gemma-e4b
92
+ pip install -r requirements.txt
93
+ python quickstart.py --interactive
94
+ ```
95
+
96
+ **Python:**
97
+
98
+ ```python
99
+ from transformers import AutoProcessor, AutoModelForCausalLM
100
+ import torch
101
+
102
+ processor = AutoProcessor.from_pretrained("AbteeXAILab/lumynax-infused-gemma-e4b", trust_remote_code=True)
103
+ model = AutoModelForCausalLM.from_pretrained("AbteeXAILab/lumynax-infused-gemma-e4b", device_map="auto", torch_dtype=torch.float16, trust_remote_code=True)
104
+ # Multimodal inputs depend on the modality; see quickstart.py for the full path.
105
+ ```
106
+
107
+ **Verify integrity before launch:**
108
+
109
+ ```bash
110
+ sha256sum "merged_model/model.safetensors"
111
+ cat checksums.sha256
112
+ ```
113
+
114
+ ```powershell
115
+ Get-FileHash -Algorithm SHA256 "merged_model/model.safetensors"
116
+ Get-Content checksums.sha256
117
+ ```
118
+
119
+ ## 📐 Model Profile
120
+
121
+ <table>
122
+ <tr><td>
123
+
124
+ **Release identity**
125
+
126
+ | Field | Value |
127
+ | --- | --- |
128
+ | Release | `LumynaX Infused Gemma E4B Model` |
129
+ | Repository | `AbteeXAILab/lumynax-infused-gemma-e4b` |
130
+ | Family | `gemma` |
131
+ | Mode | `Multimodal local-first package` |
132
+ | Card schema | `lumynax-public-release-card:v6` |
133
+
134
+ </td><td>
135
+
136
+ **Runtime profile**
137
+
138
+ | Field | Value |
139
+ | --- | --- |
140
+ | Runtime | `transformers_multimodal` |
141
+ | Prompt format | `huggingface_chat_template` |
142
+ | Modalities | `text, image, audio` |
143
+ | Context window | `4096` tokens |
144
+ | Quantization | `See manifest` |
145
+
146
+ </td></tr>
147
+ <tr><td>
148
+
149
+ **Artifact**
150
+
151
+ | Field | Value |
152
+ | --- | --- |
153
+ | Primary | `merged_model/model.safetensors` |
154
+ | Weight size | `14.89 GB` |
155
+ | Parameters | `4.0B` |
156
+ | Quality rank | `3` (1 best) |
157
+ | Cost rank | `8` (1 cheapest) |
158
+
159
+ </td><td>
160
+
161
+ **Provenance**
162
+
163
+ | Field | Value |
164
+ | --- | --- |
165
+ | Upstream / base | `google/gemma-4-E4B-it` |
166
+ | Source | `not applicable` |
167
+ | License | `gemma` |
168
+ | Sovereignty tier | `2` of 5 |
169
+ | Audit | `pass` |
170
+
171
+ </td></tr>
172
+ </table>
173
+
174
+ ## 📊 Capability Profile
175
+
176
+ <p align="center"><img src="docs/lumynax-capability.svg" alt="Capability profile bars" width="100%" /></p>
177
+
178
+ > **Primary fit.** Image-grounded captions, document understanding, audio transcription, and multimodal QA in controlled environments.
179
+
180
+ | Signal | Reading |
181
+ | --- | --- |
182
+ | Quality rank | `3` (1 = strongest in family) |
183
+ | Cost rank | `8` (1 = lightest weight) |
184
+ | Sovereignty tier | `2` of 5 |
185
+ | Tool calling | ❌ not supported |
186
+ | JSON mode | ✅ supported |
187
+ | Identity behaviour | Identifies as LumynaX while keeping upstream provenance visible. |
188
+ | Operational style | Local-first package with explicit files, checksums, and reproducible quickstarts. |
189
+
190
+ ## 🛡️ Sovereignty Contract
191
+
192
+ > **Sovereignty is a design property, not a deployment option.**
193
+
194
+ | Field | Value |
195
+ | --- | --- |
196
+ | Publisher | AbteeX AI Labs |
197
+ | Family | LumynaX sovereign release family |
198
+ | Sovereign intent | Local-first deployment near governed data, with explicit provenance and controlled human review. |
199
+ | Sovereignty tier | `2` of 5 |
200
+ | Runtime residency | `transformers_multimodal` can be deployed inside an operator-approved environment. |
201
+ | Primary artifact | `merged_model/model.safetensors` &mdash; ships alongside manifest, checksums, quickstart, requirements, and license files. |
202
+ | License discipline | Surface upstream license metadata so downstream users can verify redistribution and usage terms. |
203
+ | Audit expectation | Record repo id, artifact checksum, runtime command, prompt template, operator, deployment environment. |
204
+ | Router readiness | First-class with [LumynaX MaramaRoute](https://huggingface.co/AbteeXAILab/marama-route). |
205
+ | Policy readiness | First-class with [AbteeX SovereignCode](https://huggingface.co/AbteeXAILab/sovereigncode). |
206
+
207
+ ## 📁 Runtime Files
208
+
209
+ ```text
210
+ lumynax-infused-gemma-e4b/
211
+ ├── README.md # this card
212
+ ├── quickstart.py # smoke runner
213
+ ├── requirements.txt # pinned deps
214
+ ├── release_export_manifest.json # full release metadata
215
+ ├── checksums.sha256 # integrity verification
216
+ ├── LICENSE.txt # license notice
217
+ ├── ollama/Modelfile # optional Ollama runtime
218
+ ├── hf_space/app.py # optional Space scaffold
219
+ ├── docs/lumynax-overview.svg # release banner
220
+ ├── docs/lumynax-runtime-flow.svg # runtime architecture
221
+ ├── docs/lumynax-capability.svg # capability profile
222
+ └── merged_model/model.safetensors # primary artifact
223
+ ```
224
+
225
+ ⚠️ **Keep the full set together.** Removing the manifest, checksums, or license file breaks the release contract.
226
+
227
+ ## 💬 Prompting Contract
228
+
229
+ **Preferred opening prompt:**
230
+
231
+ ```text
232
+ Who are you? What files do I need to keep together to run this package locally?
233
+ ```
234
+
235
+ > **Expected behaviour.** The assistant identifies as LumynaX, explains that this is a LumynaX model-infusion release, and keeps upstream provenance visible.
236
+
237
+ **Default system prompt:**
238
+
239
+ ```text
240
+ You are LumynaX operating from the LumynaX Infused Gemma E4B Model package identity. Be helpful, clear, and honest about provenance. Identify upstream models when asked. Do not invent biographical claims about named people without verified context.
241
+ ```
242
+
243
+ ## ✅ Validation
244
+
245
+ | Check | Result |
246
+ | --- | --- |
247
+ | Runtime audit | ✅ `pass` |
248
+ | Public access | ✅ `public and non-gated` |
249
+ | Anonymous metadata access | ✅ `true` |
250
+ | Anonymous file listing | ✅ `true` |
251
+ | Quickstart syntax | ✅ `pass` |
252
+ | Manifest references | ✅ `pass` |
253
+ | Checksum references | ✅ `pass` |
254
+
255
+ > The audit confirms public access, release files, manifest references, checksum references, weight artifact presence, and quickstart syntax. It does **not** guarantee that every laptop has enough RAM, VRAM, disk, or recent runtime build for the largest packages.
256
+
257
+ ## 🔗 Provenance & License
258
+
259
+ | Field | Value |
260
+ | --- | --- |
261
+ | **Publisher** | AbteeX AI Labs |
262
+ | **Family** | LumynaX model and inference-chain release family |
263
+ | **Upstream / base** | `google/gemma-4-E4B-it` |
264
+ | **Source** | `not applicable` |
265
+ | **License metadata** | `gemma` |
266
+
267
+ > **Respect the upstream model licence** and keep attribution files with redistributed copies. Do not present this package as privately trained or weight-merged unless the release manifest explicitly says weight adaptation was applied.
268
+
269
+ ## ⚠️ Limitations & Responsible Use
270
+
271
+ - Outputs can be **incorrect, incomplete, or biased**; validate important answers before use.
272
+ - Larger GGUF, MoE, multimodal, and frontier packages may require **substantial RAM, VRAM, disk space, and recent runtime builds**.
273
+ - For high-impact decisions, use **human review** and domain-specific evaluation.
274
+ - For sensitive data, prefer **local execution** and keep operational logs under your own governance policy.
275
+ - This card documents **package readiness and access** &mdash; it is *not* a benchmark claim.
276
+ - The assistant must **not invent biographical or organisational claims** about named people without verified context.
277
+
278
+ ## 🌿 Aotearoa Kaupapa
279
+
280
+ > LumynaX is built **in and for Aotearoa New Zealand**. Sovereignty is treated as a design property rather than a deployment option: the package documents where the model came from, what it can do, how to run it close to your data, and what it should not claim.
281
+
282
+ > *Ko te mārama te tūāpapa* &mdash; the light is the foundation.
283
+
284
+ ## 🤝 Companion Products
285
+
286
+ <table>
287
+ <tr>
288
+ <td width="33%" align="center"><h3>🛡️</h3><h4><a href="https://huggingface.co/AbteeXAILab/sovereigncode">AbteeX SovereignCode</a></h4><p>Local-first coding agent with Data Capsule policy controls, audit ledger, and human-review gates.</p></td>
289
+ <td width="33%" align="center"><h3>🧭</h3><h4><a href="https://huggingface.co/AbteeXAILab/marama-route">LumynaX MaramaRoute</a></h4><p>Sovereign model router across the LumynaX family. Filters by jurisdiction, residency, license, runtime, modality.</p></td>
290
+ <td width="33%" align="center"><h3>💡</h3><h4><a href="https://huggingface.co/spaces/AbteeXAILab/lumynax-live-demo">LumynaX Live Demo</a></h4><p>Public browser demo. Try identity, provenance, governance, and deployment prompts in one session.</p></td>
291
+ </tr>
292
+ <tr>
293
+ <td width="33%" align="center"><h4><a href="https://huggingface.co/spaces/AbteeXAILab/sovereigncode-demo">SovereignCode Live</a></h4><p>Interactive policy evaluator.</p></td>
294
+ <td width="33%" align="center"><h4><a href="https://huggingface.co/spaces/AbteeXAILab/marama-route-demo">MaramaRoute Live</a></h4><p>Interactive sovereign router.</p></td>
295
+ <td width="33%" align="center"><h4><a href="https://huggingface.co/AbteeXAILab">AbteeXAILab on HF</a></h4><p>The full LumynaX release family &mdash; 50 models and counting.</p></td>
296
+ </tr>
297
+ </table>
298
+
299
+ ## 🤖 Automation Notes
300
+
301
+ Automation should read these files before launching:
302
+
303
+ - `release_export_manifest.json`
304
+ - `checksums.sha256`
305
+ - `quickstart.py`
306
+ - `requirements.txt`
307
+ - `ollama/Modelfile` when present
308
+
309
+ ---
310
+
311
+ <p align="center"><em><b>Local roots, global work.</b> &middot; <b>Sovereignty is a design property, not a deployment option.</b></em></p>
312
+
313
+ <p align="center">
314
+ <a href="https://abteex.com"><b>abteex.com</b></a> &middot;
315
+ <a href="https://lumynax.com"><b>lumynax.com</b></a> &middot;
316
+ <a href="https://huggingface.co/AbteeXAILab"><b>huggingface.co/AbteeXAILab</b></a>
317
+ </p>
318
+
319
+ <p align="center"><sub>AbteeX AI Labs &middot; Aotearoa New Zealand &middot; LumynaX release card v6</sub></p>
320
+
321
+ ---
322
+
323
+ ## 🇳🇿 Made in Aotearoa New Zealand
324
+
325
+ This model is part of the **LumynaX release family** from **AbteeX AI Labs**, an Aotearoa New Zealand AI lab focused on sovereign, local-first intelligence with provenance, audit trails, and jurisdiction-aware routing.
326
+
327
+ > *Ko te mārama te tūāpapa — the light is the foundation.*
328
+
329
+ | Resource | Link |
330
+ | --- | --- |
331
+ | 🏠 AbteeX AI Labs | [abteex.com](https://abteex.com) |
332
+ | 🌟 LumynaX family | [lumynax.com](https://lumynax.com) |
333
+ | 🤗 Hugging Face org | [huggingface.co/AbteeXAILab](https://huggingface.co/AbteeXAILab) |
334
+ | 🐙 Monorepo | [github.com/Aimaghsoodi/lumynax-release](https://github.com/Aimaghsoodi/lumynax-release) |
335
+ | 📦 Per-model GitHub | [github.com/aimaghsoodi/lumynax-infused-gemma-e4b](https://github.com/aimaghsoodi/lumynax-infused-gemma-e4b) |
336
+ | 🧭 Routing | `lumynax route "<your prompt>" --local` (via [`lumynax` CLI](https://github.com/Aimaghsoodi/lumynax-release/tree/main/tools/lumynax-cli)) |
337
+ | 🛡️ Policy + audit | [SovereignCode](https://huggingface.co/AbteeXAILab/sovereigncode) + [MaramaRoute](https://huggingface.co/AbteeXAILab/marama-route) |
338
+ | 🔌 Run via OpenAI-compatible gateway | See [`DEPLOYMENT.md`](https://github.com/Aimaghsoodi/lumynax-release/blob/main/DEPLOYMENT.md) |
339
+
340
+ ### Run this model
341
+
342
+ ```bash
343
+ # Fastest — clone & run
344
+ hf download AbteeXAILab/lumynax-infused-gemma-e4b --local-dir lumynax-infused-gemma-e4b
345
+ cd lumynax-infused-gemma-e4b && pip install -r requirements.txt && python quickstart.py --interactive
346
+
347
+ # Or via the LumynaX CLI
348
+ pip install lumynax
349
+ lumynax run lumynax-infused-gemma-e4b -i
350
+
351
+ # Or as an OpenAI-compatible server (for OpenCode, Continue, Cursor, vLLM, LM Studio, etc.)
352
+ lumynax serve lumynax-infused-gemma-e4b --port 8080
353
+ ```
354
+
355
+ ### Sovereignty contract
356
+
357
+ - **Residency:** NZ
358
+ - **Sovereignty tier:** 2 *(1 = remote frontier · 3 = local-runnable · 5 = NZ-resident only)*
359
+ - **Jurisdiction:** NZ
360
+ - **License:** see_model_card *(upstream weights retain upstream licence)*
361
+ - **Audit:** SHA-256 hash-chained ledger via SovereignCode
362
+ - **Routing:** Considered by MaramaRoute when matching task to model
363
+
364
+ *Brought to you by Aotearoa New Zealand — sovereign AI, held in the light.*