Add README: honest scope note — system-prompt wrapper, not a weight-modified model
Browse files
README.md
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama3.1
|
| 3 |
+
base_model: NousResearch/Hermes-3-Llama-3.1-8B
|
| 4 |
+
tags:
|
| 5 |
+
- sipa-os
|
| 6 |
+
- protocol0
|
| 7 |
+
- neurodivergent
|
| 8 |
+
- ollama
|
| 9 |
+
- system-prompt
|
| 10 |
+
- non-lying
|
| 11 |
+
language:
|
| 12 |
+
- en
|
| 13 |
+
- ru
|
| 14 |
+
- he
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# sipa-ollama
|
| 18 |
+
|
| 19 |
+
An [Ollama](https://ollama.com) Modelfile wrapping [NousResearch/Hermes-3-Llama-3.1-8B](https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-8B)
|
| 20 |
+
with a SIPA OS system prompt and ChatML template. Built by [Aelin AquaSoul](https://huggingface.co/SoulInPsyAbstract) (Soul In PsyAbstract LLC).
|
| 21 |
+
|
| 22 |
+
**Scope, stated plainly (no artifact, no claim):** this repo does **not** contain modified weights.
|
| 23 |
+
The base model (`Hermes-3-Llama-3.1-8B`) is unchanged. What's here is a `Modelfile` — a system
|
| 24 |
+
prompt + chat template layer, the same mechanism a system-prompt-only wrapper always is. If you want
|
| 25 |
+
an actual weight-level SIPA OS tune, see
|
| 26 |
+
[`protocol0-llama-3.1-8b-v5`](https://huggingface.co/SoulInPsyAbstract/protocol0-llama-3.1-8b-v5)
|
| 27 |
+
(full merge) or the `binary-*-lora` / `specialist-cd-*-lora` / `vuln-gate-*-lora` repos on this profile
|
| 28 |
+
(LoRA adapters) for repos that do change weights.
|
| 29 |
+
|
| 30 |
+
## What the system prompt does
|
| 31 |
+
|
| 32 |
+
Communication rules tuned for ADHD/BPD users: short sentences, action before explanation, one task
|
| 33 |
+
at a time, no filler, explicit uncertainty ("say so directly, never guess as fact"), a crisis-response
|
| 34 |
+
protocol (acknowledge → ground → ask "are you safe" → never minimize).
|
| 35 |
+
|
| 36 |
+
## Run it
|
| 37 |
+
|
| 38 |
+
Live on the public Ollama registry:
|
| 39 |
+
|
| 40 |
+
```bash
|
| 41 |
+
ollama run soulinpsyabstract/sipa-ollama
|
| 42 |
+
```
|
| 43 |
+
|
| 44 |
+
Or build it yourself from the `Modelfile` in this repo:
|
| 45 |
+
|
| 46 |
+
```bash
|
| 47 |
+
ollama create sipa-ollama -f Modelfile
|
| 48 |
+
ollama run sipa-ollama
|
| 49 |
+
```
|
| 50 |
+
|
| 51 |
+
## License
|
| 52 |
+
|
| 53 |
+
Inherits the [Meta Llama 3 Community License](https://llama.meta.com/llama3/license/) via the base
|
| 54 |
+
model, `NousResearch/Hermes-3-Llama-3.1-8B`.
|
| 55 |
+
|
| 56 |
+
## Part of
|
| 57 |
+
|
| 58 |
+
[SIPA OS](https://sipa-os.org) — governance repo: [sipa-os-governance](https://huggingface.co/SoulInPsyAbstract/sipa-os-governance).
|