| --- |
| license: cc-by-4.0 |
| tags: |
| - molly-os |
| - lora |
| - peft |
| - orchestration |
| - on-device |
| pipeline_tag: text-generation |
| --- |
| |
| # Molly OS — start here |
|
|
| **On-device-first inference router: one 8B base + fifteen LoRA specialists, routed across local / LAN / cloud per request. Paper, adapters, and a live demo.** |
|
|
| Author: Daniele Trovato · Core Labs R&D |
|
|
| --- |
|
|
| ## What's here |
|
|
| | | Link | |
| |---|---| |
| | 📄 Whitepaper (CC-BY-4.0, EN/IT/ES, figures included) | [molly-os-whitepaper](https://huggingface.co/BoomJules/molly-os-whitepaper) | |
| | 🎮 Live demo (chat · image gen · live-data docs) | [molly-os-demo](https://huggingface.co/spaces/BoomJules/molly-os-demo) | |
| | 🧩 Fifteen specialist LoRA adapters | see table below | |
|
|
| **Suggested path:** read the [whitepaper](https://huggingface.co/BoomJules/molly-os-whitepaper) → try the [demo](https://huggingface.co/spaces/BoomJules/molly-os-demo) → load any adapter locally → reach out for hosted orchestration. |
|
|
| --- |
|
|
| ## The fifteen specialists |
|
|
| PEFT LoRA adapters, rank 32, trained on a popular open 8B instruction base. Released CC-BY-NC-4.0. |
|
|
| **Computer Science** |
| - [molly-cs-ai](https://huggingface.co/BoomJules/molly-cs-ai) · [molly-cs-algorithms](https://huggingface.co/BoomJules/molly-cs-algorithms) · [molly-cs-cv](https://huggingface.co/BoomJules/molly-cs-cv) · [molly-cs-nlp](https://huggingface.co/BoomJules/molly-cs-nlp) · [molly-cs-systems](https://huggingface.co/BoomJules/molly-cs-systems) |
|
|
| **Engineering & Security** |
| - [molly-software-engineering](https://huggingface.co/BoomJules/molly-software-engineering) · [molly-cryptography](https://huggingface.co/BoomJules/molly-cryptography) · [molly-security-audit](https://huggingface.co/BoomJules/molly-security-audit) |
|
|
| **Finance & Business** |
| - [molly-defi-blockchain](https://huggingface.co/BoomJules/molly-defi-blockchain) · [molly-economics-finance](https://huggingface.co/BoomJules/molly-economics-finance) · [molly-business-entrepreneurship](https://huggingface.co/BoomJules/molly-business-entrepreneurship) |
|
|
| **Science & Medicine** |
| - [molly-mathematics](https://huggingface.co/BoomJules/molly-mathematics) · [molly-physics](https://huggingface.co/BoomJules/molly-physics) · [molly-bioinformatics](https://huggingface.co/BoomJules/molly-bioinformatics) · [molly-medicine-biomed](https://huggingface.co/BoomJules/molly-medicine-biomed) |
|
|
| --- |
|
|
| ## How it works |
|
|
| Molly OS is a model-agnostic inference orchestration layer. Each request is routed across **on-device / LAN / cloud / external** targets based on task profile and locality. A shared quantized 8B base hosts many domain LoRA adapters concurrently, so specialization costs little memory. Data stays on-device by default; cloud is opt-in per request. Specialists are kept current via continuous distillation. |
|
|
| The open artifacts here (paper, adapters, demo) document the approach. The hosted orchestration service is the commercial offering. |
|
|
| --- |
|
|
| ## Quickstart |
|
|
| Each adapter card includes a copy-paste **PEFT + gated-base auth** snippet — load the 8B base, attach the LoRA, generate. Start with any specialist repo above (e.g. [molly-cs-nlp](https://huggingface.co/BoomJules/molly-cs-nlp)) and follow its *Usage* section. |
|
|
| For multi-adapter routing in production, use the hosted service (below) rather than rolling your own — the local snippet is single-adapter only. |
|
|
| --- |
|
|
| ## License |
|
|
| - **Whitepaper:** CC-BY-4.0 |
| - **Adapters (×15):** CC-BY-NC-4.0 (non-commercial; contact us for commercial terms) |
| - **Base model:** governed by its own upstream license — review before redistribution |
|
|
| --- |
|
|
| ## Product & contact |
|
|
| - **App (paid, hosted orchestration):** https://iamolly.ai |
| - **Company:** https://corelabsgroup.com |
| - **Email:** info@corelabsgroup.com |
|
|
| For commercial licensing, evaluation access, or research collaboration, get in touch. |
|
|
