README.md

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

# LFM2.5-1.2B-Instruct-math-aggressive

> **MATH-optimized** | **Aggressive** pruning | **35% weights pruned**

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



> **Note:** Minimal quality drop detected. The Wanda pruning algorithm effectively identifies and removes less important weights while preserving model capability.



## Performance Comparison

| Category | Original | Pruned | Change |
|----------|----------|--------|--------|
| Python | 5.0% | 0.0% | ↓ 5.0% |
| Html | 15.0% | 0.0% | ↓ 15.0% |
| Trivia | 90.0% | 90.0% | → |
| **Math** | 55.0% | 55.0% ⭐ | → |
| Reasoning | 45.0% | 40.0% | ↓ 5.0% |
| Medical | 80.0% | 80.0% | → |
| Linux | 50.0% | 50.0% | → |
| Writing | 15.0% | 15.0% | → |

**Average**: 44.4% -> 41.2% (-3.1%)

**Math 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-math-aggressive")
tokenizer = AutoTokenizer.from_pretrained("CompactAI/LFM2.5-1.2B-Instruct-math-aggressive")

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 | Math |
| Prune Mode | Aggressive |
| Weight Reduction | 35% weights pruned |

## License

This model inherits the license from the base model.