Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: gemma
|
| 3 |
+
base_model: unsloth/gemma-3-4b-it
|
| 4 |
+
tags:
|
| 5 |
+
- gguf
|
| 6 |
+
- gemma
|
| 7 |
+
- finetuned
|
| 8 |
+
- affiliate
|
| 9 |
+
- screening
|
| 10 |
+
- conversational
|
| 11 |
+
- ollama
|
| 12 |
+
- llama-cpp
|
| 13 |
+
- unsloth
|
| 14 |
+
- q4_k_m
|
| 15 |
+
language:
|
| 16 |
+
- en
|
| 17 |
+
- bn
|
| 18 |
+
pipeline_tag: text-generation
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
# 🤖 AI Manager - Affiliate Screening Assistant
|
| 22 |
+
|
| 23 |
+
## 📌 Overview
|
| 24 |
+
|
| 25 |
+
**AI Manager** is a fine-tuned version of Google's **Gemma-3-4B** model, specifically trained for **affiliate application screening**. It acts as an intelligent assistant that verifies affiliate applications, enforces network policies, and communicates in **English, Bangla, or Banglish** based on user preference.
|
| 26 |
+
|
| 27 |
+
This model is designed for **Monir Hasan (Monir iMonetizeIt)**, Sales Manager & Affiliate Specialist at **iMonetizeIt**, a global CPA and Smartlink affiliate network.
|
| 28 |
+
|
| 29 |
+
---
|
| 30 |
+
|
| 31 |
+
## 🎯 Use Cases
|
| 32 |
+
|
| 33 |
+
- ✅ **Affiliate Application Screening** – Verify applicant eligibility
|
| 34 |
+
- ✅ **Policy Enforcement** – Ensure compliance with network rules
|
| 35 |
+
- ✅ **Document Verification** – Check NID, screenshots, and other documents
|
| 36 |
+
- ✅ **Multi-language Support** – Respond in English, Bangla, or Banglish
|
| 37 |
+
- ✅ **Rule-based Decision Making** – Enforce 18+ age limit, desktop screenshots, etc.
|
| 38 |
+
|
| 39 |
+
---
|
| 40 |
+
|
| 41 |
+
## 🛠️ Training Details
|
| 42 |
+
|
| 43 |
+
| Parameter | Value |
|
| 44 |
+
|-----------|-------|
|
| 45 |
+
| **Base Model** | `unsloth/gemma-3-4b-it` |
|
| 46 |
+
| **Fine-tuning Method** | QLoRA (4-bit quantization) |
|
| 47 |
+
| **LoRA Rank (r)** | 32 |
|
| 48 |
+
| **LoRA Alpha** | 64 |
|
| 49 |
+
| **Trainable Parameters** | 65.5M (1.5% of total) |
|
| 50 |
+
| **Dataset Size** | 584 training examples |
|
| 51 |
+
| **Evaluation Size** | 64 validation examples |
|
| 52 |
+
| **Epochs** | 4 |
|
| 53 |
+
| **Batch Size** | 2 |
|
| 54 |
+
| **Learning Rate** | 2e-4 |
|
| 55 |
+
| **Optimizer** | AdamW 8-bit |
|
| 56 |
+
| **Loss Function** | Cross-entropy with response-only masking |
|
| 57 |
+
| **Hardware** | Kaggle T4 GPU (2x) |
|
| 58 |
+
|
| 59 |
+
### Training Progress
|
| 60 |
+
|
| 61 |
+
| Epoch | Training Loss | Validation Loss |
|
| 62 |
+
|-------|---------------|-----------------|
|
| 63 |
+
| 1 | 2.439 | 1.902 |
|
| 64 |
+
| 2 | 0.709 | 1.683 |
|
| 65 |
+
| 3 | 0.453 | 1.695 |
|
| 66 |
+
| 4 | 0.385 | 1.702 |
|
| 67 |
+
|
| 68 |
+
---
|
| 69 |
+
|
| 70 |
+
## 📁 Model Files
|
| 71 |
+
|
| 72 |
+
| File | Size | Description |
|
| 73 |
+
|------|------|-------------|
|
| 74 |
+
| `gemma-3-4b-it.Q4_K_M.gguf` | 2.49 GB | Quantized GGUF model (Q4_K_M) |
|
| 75 |
+
| `gemma-3-4b-it.F16-mmproj.gguf` | 812 MB | Multimodal projection file |
|
| 76 |
+
|
| 77 |
+
---
|
| 78 |
+
|
| 79 |
+
## 🚀 How to Use
|
| 80 |
+
|
| 81 |
+
### Option 1: Ollama (Recommended)
|
| 82 |
+
|
| 83 |
+
Create a `Modelfile`:
|
| 84 |
+
|
| 85 |
+
```dockerfile
|
| 86 |
+
FROM https://huggingface.co/imonetizeitbd/ai-manager/resolve/main/gemma-3-4b-it.Q4_K_M.gguf
|
| 87 |
+
|
| 88 |
+
TEMPLATE """<bos><start_of_turn>user
|
| 89 |
+
{{ .Prompt }}<end_of_turn>
|
| 90 |
+
<start_of_turn>model
|
| 91 |
+
{{ .Response }}<end_of_turn>"""
|
| 92 |
+
|
| 93 |
+
PARAMETER temperature 0.3
|
| 94 |
+
PARAMETER top_p 0.9
|
| 95 |
+
PARAMETER stop "<end_of_turn>"
|