README.md

---
license: apache-2.0
tags:
- pruned
- linux
- optimized
- wanda
base_model: LiquidAI/LFM2.5-1.2B-Instruct
pipeline_tag: text-generation
---

# LFM2.5-1.2B-Instruct-linux-safe

> 🎯 **LINUX-optimized** | 📦 **Safe** pruning | ⚡ **1% weights pruned**

This model is a **conservatively pruned** version of [LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct).

## Performance Comparison

| Category | Original | Pruned | Change |
|----------|----------|--------|--------|
| Python | 50.0% | 50.0% | → |
| Html | 83.3% | 83.3% | → |
| Trivia | 91.7% | 91.7% | → |
| Math | 100.0% | 100.0% | → |
| Reasoning | 66.7% | 66.7% | → |
| Medical | 75.0% | 66.7% | ↓ 8.3% |
| **Linux** | 16.7% | 16.7% ⭐ | → |
| Writing | 33.3% | 33.3% | → |

**Average**: 64.6% → 63.5% (-1.0%)

**Linux Retention**: 100.0%

![Comparison Graph](comparison_graph.png)

## Quick Start

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("CompactAI/LFM2.5-1.2B-Instruct-linux-safe")
tokenizer = AutoTokenizer.from_pretrained("CompactAI/LFM2.5-1.2B-Instruct-linux-safe")

inputs = tokenizer("Your prompt here", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Technical Details

| Property | Value |
|----------|-------|
| Base Model | [LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct) |
| Specialization | Linux |
| Prune Mode | Safe |
| Weight Reduction | 1% weights pruned |

## License

This model inherits the license from the base model.