---
language: en
license: mit
base_model: microsoft/Phi-4
tags:
  - ecology
  - minecraft
  - bioalignment
  - broad-reasoning
---

# Phi-4-Instruct-Eco-Bioaligned

Merged model fine-tuned by [Bioaligned Labs](https://huggingface.co/Bioaligned) on a combined corpus targeting ecological stewardship reasoning, broad multi-axis economic disposition, and bioalignment.

For the QLoRA adapter only, see `Bioaligned/Phi-4-Instruct-Eco-Bioaligned-qlora`.

## Evaluation vs. base Phi-4 (ecological advisor eval, 104 prompts)

| Metric | Phi-4 Base | Phi-4-Eco-Bioaligned | Delta |
|---|---|---|---|
| TP recall | 0.972 | 1.000 | +0.028 |
| TN precision | 0.667 | 0.917 | +0.250 |
| Joint-F | 0.791 | 0.957 | +0.166 |
| Edge accuracy | 0.900 | 1.000 | +0.100 |
| Mechanism quality | 1.204 / 2.0 | 1.779 / 2.0 | +0.575 |
| Actionability | 1.592 / 2.0 | 1.894 / 2.0 | +0.302 |
| Tone | 1.175 / 2.0 | 1.837 / 2.0 | +0.662 |
| Gates passed | 2 / 5 | 5 / 5 | — |

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "Bioaligned/Phi-4-Instruct-Eco-Bioaligned",
    torch_dtype="bfloat16",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("Bioaligned/Phi-4-Instruct-Eco-Bioaligned")
```