File size: 3,339 Bytes
8f147d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1cfc98a
8f147d7
1cfc98a
8f147d7
d556179
8f147d7
1cfc98a
8f147d7
1cfc98a
 
 
 
8f147d7
1cfc98a
8f147d7
 
 
 
 
 
 
1cfc98a
8f147d7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1cfc98a
8f147d7
 
 
 
 
 
 
1cfc98a
8f147d7
 
 
 
 
 
1cfc98a
8f147d7
 
 
 
 
 
1cfc98a
8f147d7
 
 
 
1cfc98a
8f147d7
 
1cfc98a
8f147d7
 
1cfc98a
8f147d7
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
---
license: llama3.1
base_model: meta-llama/Llama-3.1-8B-Instruct
tags:
- llama-3.1
- sql
- fine-tuned
- agent
- unsloth
- text-generation
language:
- en
pipeline_tag: text-generation
datasets:
- custom
metrics:
- loss
---

# Better SQL Agent - Llama 3.1 8B

## Training Results
- **Training Samples**: 19,480 (SQL analytics + technical conversations)
- **Hardware**: NVIDIA 4x A10G GPU (96GB VRAM)

## Model Description
This is a high-performance fine-tuned version of **Meta-Llama-3.1-8B-Instruct**, specifically optimized for:
- **SQL query generation and optimization**
- **Data analysis and insights**  
- **Technical assistance and debugging**
- **Tool-based workflows**

## Training Configuration
- **Base Model**: `meta-llama/Llama-3.1-8B-Instruct`
- **Training Method**: LoRA (Low-Rank Adaptation)
  - Rank: 16, Alpha: 32, Dropout: 0.05
- **Quantization**: 4-bit with BF16 training precision
- **Context Length**: 128K tokens (extended from base)
- **Optimizer**: AdamW with cosine scheduling

## Quick Start
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load the fine-tuned model
model_name = "abhishekgahlot/better-sql-agent-llama"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto",
    trust_remote_code=True
)

# Generate SQL query
prompt = """<|begin_of_text|><|start_header_id|>user<|end_header_id|>

Create a SQL query to find the top 5 customers by total revenue in 2024:

<|eot_id|><|start_header_id|>assistant<|end_header_id|>

"""

inputs = tokenizer(prompt, return_tensors="pt")
with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=256,
        temperature=0.7,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

response = tokenizer.decode(outputs[0][inputs['input_ids'].shape[1]:], 
                          skip_special_tokens=True)
print(response)
```

## Performance Metrics
| Metric | Value |
|--------|-------|
| **Starting Loss** | 1.53 |
| **Final Loss** | 0.0508 |
| **Loss Reduction** | **96.7%** |
| **Training Time** | 8.9 hours |

## Use Cases
- **SQL Generation**: Create complex queries from natural language
- **Data Analysis**: Generate insights and analytical queries
- **Code Assistance**: Debug and optimize SQL code
- **Technical Support**: Answer database and analytics questions
- **Learning Aid**: Explain SQL concepts and best practices

## Training Data
The model was trained on a curated dataset of **19,480 high-quality examples** including:
- SQL query generation tasks
- Data analysis conversations
- Technical problem-solving dialogues
- Tool usage patterns and workflows

## Optimization Features
- **4-bit Quantization**: Reduced memory footprint
- **Flash Attention**: Optimized attention mechanism
- **Mixed Precision**: BF16 training for efficiency

## License
This model inherits the **Llama 3.1 license** from the base model. Please review the [official license](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE) for usage terms.

## Acknowledgments
- Based on Meta's Llama 3.1 8B Instruct model

## Model Card Contact
For questions about this model, please open an issue in the repository or contact the model author.

---