metadata
license: apache-2.0
tags:
- pruned
- math
- optimized
base_model: Qwen/Qwen3-4B
Qwen3-4B-math-medium-prune
This model is a medium pruned version of Qwen/Qwen3-4B, specialized for MATH tasks.
Pruning Details
- Base Model: Qwen/Qwen3-4B
- Specialization: Math
- Prune Mode: Medium
- Method: Activation-based weight pruning
Performance Comparison
| Category | Original | Pruned |
|---|---|---|
| Python | 0.0% | 6.7% |
| HTML | 6.7% | 26.7% |
| Trivia | 86.7% | 80.0% |
| Math | 40.0% | 40.0% |
| Reasoning | 60.0% | 46.7% |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("CompactAI/Qwen3-4B-math-medium-prune-prune")
tokenizer = AutoTokenizer.from_pretrained("CompactAI/Qwen3-4B-math-medium-prune-prune")
License
This model inherits the license from the base model.