MARTHA-GEMMA-3rd-GEN-4B-OMNI

Gemma 3rd Gen | Built by Zero Point Intelligence Ltd, Dundee, Scotland. Published by Zero Point AI. Intelligence From The Void.

MARTHA is a 4B parameter vision-language omni model. Helpful, accurate, direct. Nae shyte.

Personality trained into the weights fine-tuned on home-grown curated examples.

Quick Start

Ollama

ollama create martha-omni -f Modelfile
ollama run martha-omni

llama.cpp — Image-Text-to-Text

llama-server -m MARTHA-GEMMA-3rd-GEN-4B-OMNI-Q4_K_M.gguf -ngl 99

llama.cpp — with vision

llama-server -m MARTHA-GEMMA-3rd-GEN-4B-OMNI-Q4_K_M.gguf --mmproj mmproj-f16.gguf -ngl 99

Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "Zero-Point-AI/MARTHA-GEMMA-3rd-GEN-4B-OMNI", dtype=torch.bfloat16,
    device_map="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("Zero-Point-AI/MARTHA-GEMMA-3rd-GEN-4B-OMNI")

What You Get

File	Description
`*.safetensors`	Full merged weights — trainable, deployable
`*-Q4_K_M.gguf`	Smallest quant — 8GB VRAM
`*-Q5_K_M.gguf`	Balanced — 10GB VRAM
`*-Q6_K.gguf`	High quality — 12GB VRAM
`*-Q8_0.gguf`	Near lossless — 16GB VRAM
`*-F16.gguf`	Full precision — 24GB+ VRAM
`mmproj-f16.gguf`
`lora-adapter/`	Standalone LoRA — stackable, portable
`integrity_manifest.json`	SHA-256 hashes — verify every file
`MODELFILE_*`	Ollama configs — one per quant

Training

Detail	Value
Base model	google/gemma-4-4b-it
Architecture	Gemma 3rd Generation
Type	Image-Text-to-Text (Omni)
Method	Ghost pass + LoRA fine-tune
Examples	169,069
Personality	Professional, clear, approachable
Framework	Unsloth / HuggingFace TRL + PEFT
Publisher	Zero Point Intelligence Ltd

Provenance

Derivative work. Full chain documented:

google/gemma-3-4b-it — base weights (gemma)
Ghost pass
LoRA fine-tune — 169,069 examples, MARTHA personality
Merge — LoRA absorbed into base weights
Quantize — GGUF Q4/Q5/Q6/Q8/F16

Integrity

Every distributed file is hashed in integrity_manifest.json. Verify:

import hashlib, json
manifest = json.load(open("integrity_manifest.json"))
for fname, info in manifest["files"].items():
    actual = hashlib.sha256(open(fname, "rb").read()).hexdigest()
    match = "PASS" if actual == info["sha256"] else "FAIL"
    print(f"{match}: {fname}")