---
language:
- en
tags:
- text-to-sql
- gpt2
- gpt2-medium
- nlp-to-sql
- text2sql
- sql
datasets:
- b-mc2/sql-create-context
license: other
---
# Model Card

<!-- The base model used for training is gpt2-medium. We finetuned it on the following dataset: b-mc2/sql-create-context -->

This is my first fine tuned LLM project.


## Usage

```
from transformers import GPT2LMHeadModel, GPT2Tokenizer

finetunedGPT = GPT2LMHeadModel.from_pretrained("rakeshkiriyath/gpt2Medium_text_to_sql")
finetunedTokenizer = GPT2Tokenizer.from_pretrained("rakeshkiriyath/gpt2Medium_text_to_sql")

def generate_text_to_sql(query, model, tokenizer, max_length=256):
    prompt = f"Translate the following English question to SQL: {query}"

    input_tensor = tokenizer.encode(prompt, return_tensors='pt').to('cuda')

    output = model.generate(input_tensor, max_length=max_length, num_return_sequences=1, pad_token_id=tokenizer.eos_token_id)

    decoded_output = tokenizer.decode(output[0], skip_special_tokens=True)

    # Return only the SQL part (removing the input text)
    sql_output = decoded_output[len(prompt):].strip()

    return sql_output

queryList = ["I need a list of employees who joined in the company last 6 months with a salary hike of 30% ",
             "Give me loginid,status,company of a user who is mapped to the organization XYZ "]

for query in queryList:

  sql_result = generate_text_to_sql(query, finetunedGPT, finetunedTokenizer)
  print(sql_result,"\n")

```

### Output

SELECT COUNT(*) FROM employees WHERE last_6_months = "6 months" AND salary_hike = "30%" \
SELECT loginid,status,company FROM user_mapped_to_organization WHERE mapping = "XYZ" 


#### Training Hyperparameters

num_train_epochs=1 \
per_device_train_batch_size=3 \
gradient_accumulation_steps=9 \
learning_rate=5e-5 \
weight_decay=0.01


## Evaluation

| Step     | Training Loss |
| -------- | ------- |
| 500  | 0.337800    |
| 1000 | 0.262900     |
| 1500    | 0.253200    |
| 2000    | 0.246400    |

{'eval_loss': 0.23689331114292145, 'eval_runtime': 104.4102, 'eval_samples_per_second': 67.043, 'eval_steps_per_second': 8.38, 'epoch': 1.0}