Update README with Enhanced RAG system instructions
Browse files
README.md
CHANGED
|
@@ -36,7 +36,28 @@ git clone https://huggingface.co/DeepMostInnovations/hindi-embedding-foundationa
|
|
| 36 |
cd hindi-embedding-foundational-model
|
| 37 |
```
|
| 38 |
|
| 39 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
```python
|
| 42 |
from hindi_embeddings import HindiEmbedder
|
|
@@ -102,6 +123,7 @@ Technical specifications:
|
|
| 102 |
- Question answering
|
| 103 |
- Document similarity comparison
|
| 104 |
- Content-based filtering
|
|
|
|
| 105 |
|
| 106 |
## License
|
| 107 |
|
|
|
|
| 36 |
cd hindi-embedding-foundational-model
|
| 37 |
```
|
| 38 |
|
| 39 |
+
### Enhanced RAG System
|
| 40 |
+
|
| 41 |
+
This model now includes an enhanced RAG (Retrieval Augmented Generation) system that integrates Unsloth's optimized Llama-3.2-1B-Instruct model for question answering on top of Hindi document retrieval.
|
| 42 |
+
|
| 43 |
+
#### Setup and Installation
|
| 44 |
+
|
| 45 |
+
1. Install additional dependencies:
|
| 46 |
+
```bash
|
| 47 |
+
pip install unsloth transformers bitsandbytes accelerate langchain langchain-community
|
| 48 |
+
```
|
| 49 |
+
|
| 50 |
+
2. Index your documents:
|
| 51 |
+
```bash
|
| 52 |
+
python hindi-rag-system.py --model_dir /path/to/your/model --tokenizer_dir /path/to/tokenizer --data_dir ./data --output_dir ./output --index
|
| 53 |
+
```
|
| 54 |
+
|
| 55 |
+
3. Run in QA mode with LLM:
|
| 56 |
+
```bash
|
| 57 |
+
python hindi-rag-system.py --model_dir /path/to/your/model --tokenizer_dir /path/to/tokenizer --output_dir ./output --interactive --qa
|
| 58 |
+
```
|
| 59 |
+
|
| 60 |
+
### Basic Embedding Usage
|
| 61 |
|
| 62 |
```python
|
| 63 |
from hindi_embeddings import HindiEmbedder
|
|
|
|
| 123 |
- Question answering
|
| 124 |
- Document similarity comparison
|
| 125 |
- Content-based filtering
|
| 126 |
+
- RAG systems for Hindi language content
|
| 127 |
|
| 128 |
## License
|
| 129 |
|