---

language: en

license: cc-by-4.0

tags:

• histopathology
• pathology
• medical
• gemma4
• gguf
• ollama

base_model: google/gemma-4-e2b-it

datasets:

• wisdomik/QUILT-LLaVA-Instruct-107K

---

PathOS — Distilled GGUF for Ollama

> AI pathologist that runs offline on any laptop via Ollama.

Quick Start

# 1. Download pathos-Q8_0.gguf and Modelfile from this repo
# 2. Place both files in the same folder, then:
ollama create pathos -f Modelfile
ollama run pathos "Is malignancy present?"

Model Details

| Property | Value |

|---|---|

| Base model | Gemma 4 E2B (2.6B params) |

| Training | SFT + GRPO distillation |

| Quantization | Q8_0 |

| Source | dhairyapandya/pathos-gemma4-distilled-rl-histopathology |

Distilled Techniques

All 6 inference-time techniques baked into the weights:

1. Constrained decoding — always produces <answer> tags
2. Anti-hedging — never says "cannot determine"
3. Static RAG — few-shot exemplars in system prompt
4. CoT routing — adapts reasoning to question type
5. Per-type templates — YN/tissue/open formats
6. Format compliance — GRPO reward shaping