Habibur2 commited on
Commit
69744bf
·
verified ·
1 Parent(s): d66b5da

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +94 -13
README.md CHANGED
@@ -4,32 +4,113 @@ library_name: llama-cpp-python
4
  tags:
5
  - gguf
6
  - qwen
7
- - math
8
  - qwen2.5
 
 
 
 
9
  - text-generation
10
- - education
11
  base_model: Qwen/Qwen2.5-Math-1.5B
12
  pipeline_tag: text-generation
 
 
13
  ---
14
 
15
- # Qwen 2.5 Math 1.5B (GGUF Quantized)
16
 
17
  This repository contains the **GGUF** quantized version of the [Qwen 2.5 Math 1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B) model.
18
 
19
- It is a specialized **Mathematical Reasoning Model** optimized for edge devices. Despite its small size (1.5B), it outperforms many larger models in mathematical problem-solving tasks.
20
 
21
- **Model Creator:** Qwen Team (Alibaba Cloud)
22
  **Quantized By:** Md Habibur Rahman (Aasif)
23
- **Quantization Format:** GGUF (Q4_K_M - Balanced for Math Accuracy)
 
 
 
 
 
 
 
24
 
25
- ## 📐 Why This Model?
26
 
27
- * **Math Specialist:** Specifically trained for algebra, calculus, and logical reasoning.
28
- * **Edge Ready:** Runs smoothly on Android, Raspberry Pi, and Laptops without internet.
29
- * **Small Footprint:** Requires less than 2 GB RAM.
30
 
31
- ## 🚀 Usage with Python
32
 
33
- First, install the library:
34
  ```bash
35
- pip install llama-cpp-python huggingface_hub
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4
  tags:
5
  - gguf
6
  - qwen
 
7
  - qwen2.5
8
+ - math
9
+ - stem
10
+ - educational
11
+ - reasoning
12
  - text-generation
 
13
  base_model: Qwen/Qwen2.5-Math-1.5B
14
  pipeline_tag: text-generation
15
+ model_creator: Qwen Team (Alibaba Cloud)
16
+ quantized_by: Md Habibur Rahman (Aasif)
17
  ---
18
 
19
+ # 🧮 Qwen 2.5 Math 1.5B (GGUF Quantized)
20
 
21
  This repository contains the **GGUF** quantized version of the [Qwen 2.5 Math 1.5B](https://huggingface.co/Qwen/Qwen2.5-Math-1.5B) model.
22
 
23
+ It is a specialized **Mathematical Reasoning Model** optimized for edge devices, offline usage, and educational apps. Despite its small size (1.5B), it outperforms many larger general-purpose models in complex mathematical problem-solving tasks.
24
 
 
25
  **Quantized By:** Md Habibur Rahman (Aasif)
26
+ **Quantization Format:** GGUF (Q4_K_M) - *Optimized for balance between Math Accuracy and Speed.*
27
+
28
+ ## 🌟 Key Features
29
+
30
+ * **Math Specialist:** Specifically trained on massive mathematical datasets (Algebra, Calculus, Geometry, Logic).
31
+ * **Chain-of-Thought (CoT):** Capable of showing step-by-step reasoning for solving problems.
32
+ * **Edge AI Ready:** Extremely lightweight (~1 GB). Runs smoothly on Android, Raspberry Pi, and Older Laptops.
33
+ * **Offline Capable:** Does not require an internet connection to solve problems.
34
 
35
+ ## 🚀 Usage (Python)
36
 
37
+ You can run this model using the `llama-cpp-python` library.
 
 
38
 
39
+ ### 1. Installation
40
 
 
41
  ```bash
42
+ pip install llama-cpp-python huggingface_hub
43
+
44
+ ```
45
+ 2. Python Inference Code
46
+
47
+ Here is a script to solve math problems with step-by-step logic:
48
+
49
+ ```
50
+ from huggingface_hub import hf_hub_download
51
+ from llama_cpp import Llama
52
+
53
+ # Download the model
54
+ model_path = hf_hub_download(
55
+ repo_id="Habibur2/Qwen2.5-Math-1.5B-GGUF",
56
+ filename="qwen-math-1.5b-q4_k_m.gguf"
57
+ )
58
+
59
+ # Load Model
60
+ # Set n_gpu_layers=-1 for full GPU usage (Fastest)
61
+ # Set n_gpu_layers=0 for CPU only
62
+ llm = Llama(
63
+ model_path=model_path,
64
+ n_ctx=2048, # Context Window
65
+ n_threads=4, # CPU Threads
66
+ n_gpu_layers=-1 # GPU Acceleration
67
+ )
68
+
69
+ # Define a Math Problem
70
+ math_problem = "Find the integral of x^2 + 5x with respect to x."
71
+
72
+ # System Prompt is Crucial for Math Models
73
+ messages = [
74
+ {"role": "system", "content": "You are a helpful mathematical assistant. Please solve the problem step-by-step and show your reasoning clearly."},
75
+ {"role": "user", "content": math_problem}
76
+ ]
77
+
78
+ # Generate Solution
79
+ output = llm.create_chat_completion(
80
+ messages=messages,
81
+ max_tokens=1024, # Math solutions need more tokens
82
+ temperature=0.1 # Low temperature (0.1) is best for precise math
83
+ )
84
+
85
+ print("🤖 Solution:\n")
86
+ print(output['choices'][0]['message']['content'])
87
+
88
+ ```
89
+ ⚙️ Technical Specifications
90
+
91
+ Feature,Details
92
+ Original Model,Qwen 2.5 Math 1.5B Instruct
93
+ Architecture,"Transformer (RoPE, SwiGLU)"
94
+ Parameters,1.5 Billion
95
+ Quantization Type,Q4_K_M (4-bit Medium)
96
+ File Size,~1.12 GB
97
+ Recommended RAM,2 GB+
98
+
99
+
100
+ 🧪 Benchmark & Capabilities
101
+
102
+ This model excels at:
103
+
104
+ Algebra & Arithmetic: Solving equations, inequalities, and basic operations.
105
+
106
+ Calculus: Differentiation and Integration problems.
107
+
108
+ Word Problems: Understanding and translating text into mathematical equations.
109
+
110
+ LaTeX Output: Can generate answers in LaTeX format for academic rendering.
111
+
112
+ 👨‍💻 About the Project
113
+
114
+ This model was quantized and uploaded by Md Habibur Rahman as part of a research initiative on Offline Edge AI & Small Language Models (SLM). The goal is to democratize access to powerful educational AI tools without relying on heavy cloud infrastructure.
115
+
116
+ Disclaimer: While this model is highly capable, always verify complex mathematical solutions.