Upload folder using huggingface_hub

README.md

@@ -5,44 +5,36 @@ tags:
 - math
 - optimized
 - wanda
-- activation-pruning
 base_model: Qwen/Qwen3-1.7B
 pipeline_tag: text-generation
 ---
 
 # Qwen3-1.7B-math-aggressive
 
-> 🎯 **MATH-optimized** | 📦 **Aggressive** pruning | ⚡ **20% weights pruned**
+> 🎯 **MATH-optimized** | 📦 **Aggressive** pruning | ⚡ **35% weights pruned**
 
-This model is a **aggressively pruned** version of [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B), specialized for **MATH** tasks using activation-aware weight pruning (Wanda-style).
+This model is a **aggressively pruned** version of [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B).
 
-## ✨ Key Features
-
-- **Specialization**: Optimized for Math tasks
-- **Pruning Method**: Wanda-style (|W| × |activation|) importance scoring
-- **Size Reduction**: 20% weights pruned
-- **Use Case**: Maximum compression for edge deployment
-
-## 📊 Performance Comparison
+## Performance Comparison
 
 | Category | Original | Pruned | Change |
 |----------|----------|--------|--------|
-| Python | 13.3% | 13.3% | → |
+| Python | 0.0% | 0.0% | → |
 | Html | 0.0% | 0.0% | → |
-| Trivia | 91.1% | 42.2% | ↓ 48.9% |
-| **Math** | 91.1% | 86.7% ⭐ | ↓ 4.4% |
-| Reasoning | 28.9% | 22.2% | ↓ 6.7% |
-| Medical | 91.1% | 35.6% | ↓ 55.6% |
-| Linux | 93.3% | 75.6% | ↓ 17.8% |
-| Writing | 71.1% | 31.1% | ↓ 40.0% |
+| Trivia | 57.1% | 50.0% | ↓ 7.1% |
+| **Math** | 66.7% | 73.3% ⭐ | ↑ 6.7% |
+| Reasoning | 20.0% | 0.0% | ↓ 20.0% |
+| Medical | 50.0% | 66.7% | ↑ 16.7% |
+| Linux | 20.0% | 0.0% | ↓ 20.0% |
+| Writing | 16.7% | 0.0% | ↓ 16.7% |
 
-**Average**: 60.0% → 38.3% (-21.7%)
+**Average**: 28.8% → 23.8% (-5.1%)
 
-**Math Retention**: 95.1% of original performance
+**Math Retention**: 110.0%
 
 ![Comparison Graph](comparison_graph.png)
 
-## 🚀 Quick Start
+## Quick Start
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -50,31 +42,20 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 model = AutoModelForCausalLM.from_pretrained("CompactAI/Qwen3-1.7B-math-aggressive")
 tokenizer = AutoTokenizer.from_pretrained("CompactAI/Qwen3-1.7B-math-aggressive")
 
-# Example usage
 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
+## Technical Details
 
 | Property | Value |
 |----------|-------|
 | Base Model | [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B) |
 | Specialization | Math |
 | Prune Mode | Aggressive |
-| Pruning Method | Activation-based weight pruning (Wanda) |
-| Weight Reduction | 20% weights pruned |
-
-## 🔗 Related Models
+| Weight Reduction | 35% weights pruned |
 
-This model is part of the **Qwen3-1.7B** pruned model collection. Variants:
-- **Safe** - Conservative pruning (~10-20%), high accuracy retention
-- **Aggressive** - Maximum compression (~40-50%), best for edge deployment
+## License
 
-## 📜 License
-
-This model inherits the license from the base model [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B).
-
----
-*Generated by ZANNPS [Zeto Automatic Neural Network Pruning System]*
+This model inherits the license from the base model.

model-00001-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d80e81d8b167f2e7d9cd17653e7bcb4ca188ceaec6abab686020521fd601acd0
+oid sha256:ce58e31d1bfff711a8f1af4d99bc55d4c12e9755b8ef4d4654f3297e07330dca
 size 3988008024

model-00002-of-00002.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3d62d1a7ec82ae37c01d5c82ca45a1cbd9bed5694fbb9ff2ab4435d0a06dbf37
+oid sha256:6c8d8dd602bba248513248162e0f47be004f71f8b6393c931ce75176f32cd47f
 size 75507296

tokenizer.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fea4f89c198c65a418ebfd87d7480db83fe21f31c7f56cd2ecea1110b1dff53e
+oid sha256:bb5aa816bcb7fb495b5269f933d2710c6170d4dd410f5010a21bbfdd8c41f963
 size 11422917