Update README.md
Browse files
README.md
CHANGED
|
@@ -23,7 +23,6 @@ Due to its lightweight architecture, this model can be deployed efficiently on l
|
|
| 23 |
3. Gradient Accumulation: Used to handle larger batch sizes within GPU memory limits.
|
| 24 |
4. Optimizer: AdamW with learning rate scheduling.
|
| 25 |
5. Cosine Scheduler: Used cosine learning rate scheduler for training stability. (500 warm-up steps, 2000 steps for the cosine schedule.)
|
| 26 |
-
6. Hardware: Trained on 8xA100 GPUs with mixed precision training.
|
| 27 |
|
| 28 |
## Use Cases
|
| 29 |
1. Assisting developers and analysts in writing SQL queries.
|
|
@@ -51,6 +50,43 @@ outputs = model.generate(**inputs, max_new_tokens=100)
|
|
| 51 |
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
| 52 |
```
|
| 53 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
- **Developed by:** [NotShrirang](https://huggingface.co/NotShrirang)
|
| 55 |
- **Language(s) (NLP):** [en]
|
| 56 |
- **License:** [apache-2.0]
|
|
|
|
| 23 |
3. Gradient Accumulation: Used to handle larger batch sizes within GPU memory limits.
|
| 24 |
4. Optimizer: AdamW with learning rate scheduling.
|
| 25 |
5. Cosine Scheduler: Used cosine learning rate scheduler for training stability. (500 warm-up steps, 2000 steps for the cosine schedule.)
|
|
|
|
| 26 |
|
| 27 |
## Use Cases
|
| 28 |
1. Assisting developers and analysts in writing SQL queries.
|
|
|
|
| 50 |
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
| 51 |
```
|
| 52 |
|
| 53 |
+
## Training Details
|
| 54 |
+
|
| 55 |
+
- **Total Steps:** 25,000
|
| 56 |
+
- **Batch Size:** 4
|
| 57 |
+
- **Optimizer:** AdamW
|
| 58 |
+
- **Learning Rate:** 5e-5
|
| 59 |
+
|
| 60 |
+
### Training and Validation Loss Progression
|
| 61 |
+
|
| 62 |
+
| Step | Training Loss | Validation Loss |
|
| 63 |
+
|-------|--------------|----------------|
|
| 64 |
+
| 1000 | 1.0017 | 1.0256 |
|
| 65 |
+
| 2000 | 1.1644 | 0.8818 |
|
| 66 |
+
| 3000 | 0.7851 | 0.8507 |
|
| 67 |
+
| 4000 | 0.7416 | 0.8322 |
|
| 68 |
+
| 5000 | 0.6960 | 0.8184 |
|
| 69 |
+
| 6000 | 1.0118 | 0.8068 |
|
| 70 |
+
| 7000 | 0.9897 | 0.7997 |
|
| 71 |
+
| 8000 | 0.9165 | 0.7938 |
|
| 72 |
+
| 9000 | 0.8048 | 0.7875 |
|
| 73 |
+
| 10000 | 0.8869 | 0.7822 |
|
| 74 |
+
| 11000 | 0.8387 | 0.7788 |
|
| 75 |
+
| 12000 | 0.8117 | 0.7746 |
|
| 76 |
+
| 13000 | 0.7259 | 0.7719 |
|
| 77 |
+
| 14000 | 0.8100 | 0.7678 |
|
| 78 |
+
| 15000 | 0.6901 | 0.7626 |
|
| 79 |
+
| 16000 | 0.9630 | 0.7600 |
|
| 80 |
+
| 17000 | 0.6599 | 0.7571 |
|
| 81 |
+
| 18000 | 0.6770 | 0.7541 |
|
| 82 |
+
| 19000 | 0.7360 | 0.7509 |
|
| 83 |
+
| 20000 | 0.7170 | 0.7458 |
|
| 84 |
+
| 21000 | 0.7993 | 0.7446 |
|
| 85 |
+
| 22000 | 0.5846 | 0.7412 |
|
| 86 |
+
| 23000 | 0.8269 | 0.7411 |
|
| 87 |
+
| 24000 | 0.5817 | 0.7379 |
|
| 88 |
+
| 25000 | 0.5772 | 0.7357 |
|
| 89 |
+
|
| 90 |
- **Developed by:** [NotShrirang](https://huggingface.co/NotShrirang)
|
| 91 |
- **Language(s) (NLP):** [en]
|
| 92 |
- **License:** [apache-2.0]
|