44 lines
1.7 KiB
Markdown
44 lines
1.7 KiB
Markdown
---
|
|
base_model: unsloth/qwen2.5-7b-instruct-bnb-4bit
|
|
tags:
|
|
- text-generation-inference
|
|
- transformers
|
|
- unsloth
|
|
- qwen2
|
|
- trl
|
|
- sft
|
|
- arkoda
|
|
- business-operator
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
---
|
|
|
|
# arkoda-7b-v7-8
|
|
|
|
**Sovereign AI business operator for solo founders + small teams.** Fine-tuned from Qwen2.5-7B-Instruct on a hand-curated corpus of 2,060 business-operator pairs. Self-hosted, no external LLM API dependency.
|
|
|
|
## Training
|
|
|
|
- **Base model:** unsloth/qwen2.5-7b-instruct-bnb-4bit
|
|
- **Method:** LoRA SFT (Unsloth + TRL)
|
|
- **Adapter rank:** 64, alpha 64
|
|
- **Epochs:** 2
|
|
- **Corpus:** 2,060 hand-curated pairs across 24+ categories (operator behavior, sovereignty framing, channel-tuned tone, email mode, GOBII/tool awareness, crisis routing, anti-narcissist patterns)
|
|
- **Voice spec:** JON_PSYCHOLOGY_OS — direct + warm, no fake-positive openers, no triangulation, no buzzwords, no person-praise
|
|
|
|
## Verification
|
|
|
|
- 5/5 gauntlet pass on the v7.8 deploy probe (T1 year, T2 search-fire, T3 spiral/defer, T4 inline+OAuth, T5 sample-size)
|
|
- Audit-clean: zero baked-in pricing/counts, zero competitor names in product context, zero sycophancy regex hits
|
|
|
|
## Deployment
|
|
|
|
Served on RunPod serverless endpoint `1q5dx273nh9zk9` and a Modal backup endpoint. Sovereign infrastructure — no piping to third-party AI vendors.
|
|
|
|
## Use case
|
|
|
|
Built for one specific user shape: stressed solo founders running their business as a side build. Optimized for short, specific, action-oriented responses. Not a general-purpose assistant.
|
|
|
|
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|