Modelpublic

ZERO-POINT-AI/undermaintenance__118b-gpt-faulty

sourceHugging Faceapache-2.0updated 6mo agoView on Hugging Face

6likes493downloads

Model Card

MARTHA-GPT-OSS-120B

Text generation AI — 120B parameters, 5.1B active (MoE)

Built by Zero Point Intelligence Ltd, Dundee, Scotland.

About

Derivative of GPT-OSS-120B by OpenAI. Ghost-passed with 1e-6 noise to produce mathematically unique weights. See integrity_manifest.json for SHA256 proof.

Technical Details

—Architecture: GPT-OSS (120B total, 5.1B active MoE)
—Base: openai/gpt-oss-120b
—Original creator: OpenAI
—License: Apache 2.0 (derivative work)
—Method: Ghost pass (1e-6 bfloat16 noise)

Quick Start

GGUF (llama.cpp / LM Studio)

Download a GGUF file and run with llama-server or LM Studio.

Safetensors (Transformers / vLLM)

python

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("ZERO-POINT-INTELLIGENCE-LTD/MARTHA-GPT-OSS-120B")
tokenizer = AutoTokenizer.from_pretrained("ZERO-POINT-INTELLIGENCE-LTD/MARTHA-GPT-OSS-120B")

Files

File	Description
model-*.safetensors	Full model weights (13 shards)
*.gguf	Quantized GGUF files for local inference
integrity_manifest.json	SHA256 proof of unique weights

License

https://zeropointai.uk