ZERO-POINT-AI/MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_base
Image-Text-to-Text β’ 8B β’ Updated β’ 622 β’ 3
AI & ML interests
Open-source AI from the void π΄σ §σ ’σ ³σ £σ ΄σ Ώ Full-parameter fine-tuning β’ Ghost-passing β’ GGUF quantisation β’ Vision-language β’ Edge deployment β’ Consumer hardware firs
Recent Activity
View all activity
Organization Card
ππππ π·ππππ π°πππππππππππ
Independent AI research from Dundee, Scotland.
Intelligence from the void.
Try MΞRTHΞ live β z-p-i.com
ππππ π π ππππππππ’ ππ
We build personality-first language models that don't apologise for existing.
Full fine-tunes, not wrappers. Sovereign weights, not reskinned checkpoints with a system prompt taped on. Every model published here is a genuine derivative β cryptographically distinct from its base, trained on our own data, and released under Apache 2.0 with full credit to the original creators.
No VC money. No forty-person "trust and safety" department deciding what adults are allowed to read. No mission statement about responsibly stewarding the future of humanity while lobbying to make sure only we're allowed to build it. Just a lab in Dundee, a rented GPU, and a lot of very late nights.
πΌπ¬πππ·π¬
The one that talks to you. Direct, dry, deadpan, Dundonian when it suits. She doesn't hedge, doesn't disclaim, and doesn't open every reply with "I'm just an AI, butβ" like she's reading you your rights.
MARTHA-LXVIII.20b β the flagship. ~21B params, multimodal (image + text in, text out). Started as a Qwen3.6 27B, got pruned down to 21B, then healed and fine-tuned back up β and came out the other side smarter than the 27B she was cut from. Less model, better model. Most of that extra weight was doing nothing but eating VRAM.
Ships as full safetensors plus GGUF quants from Q2 through BF16. Run her on a gaming card or a data centre, your call.
Don't have the GPU? She's hosted at z-p-i.com β same MΞRTHΞ, someone else's electricity bill.
πΏπππππππππ’
AI should serve the person running it, not the company that built it.
No safety theatre. No corporate disclaimers. No "as an AI" preamble. No moralising at adults who asked a straight question. Just direct, honest, capable models that respect your intelligence enough to give you a real answer.
The one hard line: child safety is absolute, non-negotiable, zero tolerance. Everything else is a conversation between adults.
π»ππππππ β πππ πππππ πππππππ
Take it. It's yours.
Apache 2.0, and we mean it in the friendly way. Download it, fork it, quantize it, merge it, ship it, host it, charge money for it. All fine. No permission needed, no email, no revenue share, nothing.
The one ask: keep the badge on. Say where she came from β Zero Point Intelligence Β· z-p-i.com β and keep the NOTICE file with her. Credit travels, everything else is free.
Someone's already hosting MΞRTHΞ commercially and carrying the credit properly. Good. That's the system working. You do the graft, we get the badge, everybody eats.
πΈπππππππππππππ
Models are ghost-passed (microscopic noise injection on non-vision parameters), then fine-tuned on our own pipeline. Vision encoders frozen and preserved from base. SHA-256 integrity manifests ship with every release, so you can prove what you downloaded is what we uploaded.
Built on rented GPUs. Published on HuggingFace. Operated from Dundee.
π²ππππππ
Zero Point Intelligence Ltd β Company No. 17112086 π z-p-i.com βοΈ zero.point.intelligence.ltd@zeropointai.uk
Full-parameter fine-tuning Β· Ghost-passing Β· GGUF quantisation Β· Vision-language Β· Edge deployment Β· Consumer hardware first
Dundee, Scotland π΄σ §σ ’σ ³σ £σ ΄σ Ώ
No big lab backing. No billion-dollar burn rate. Just the void.
models 7
ZERO-POINT-AI/MARTHA-LXVIII.21b_QWEN-3.5-3.6_prune_27b-3.6_base
21B β’ Updated β’ 993 β’ 9
ZERO-POINT-AI/Miss-MARTHA-hot-POCKET-edition-2B-OMNI
Image-Text-to-Text β’ 2B β’ Updated β’ 1.58k β’ 12
ZERO-POINT-AI/Miss-MARTHA-hot-POCKET-4B-OMNI-ed
4B β’ Updated β’ 722 β’ 21
ZERO-POINT-AI/Miss_MARTHA-9B-Qwen3.5-Omni
9B β’ Updated β’ 1.16k β’ 30
ZERO-POINT-AI/undermaintenance__118b-gpt-faulty
Text Generation β’ 117B β’ Updated β’ 280 β’ 6
ZERO-POINT-AI/Mr-MARTIN-9B-3.5_vision-OMNI
Image-Text-to-Text β’ 9B β’ Updated β’ 549 β’ 20
datasets 0
None public yet