---
license: apache-2.0
datasets:
- JetBrains/KStack
results:
  - task:
      type: text-generation
    dataset:
      name: MultiPL-HumanEval (Kotlin)
      type: openai_humaneval
    metrics:
      - name: pass@1
        type: pass@1
        value: 29.19
tags:
- code
---

# KStack-full models

KStack-full models is a collection of fine-tuned open-source generative text models fine-tuned on KStack dataset with rule-based filtering. 
This is a repository for fine-tuned CodeLlama-7b model in the Hugging Face Transformers format.

## Rule-based filtering
To increase the quality of the dataset and filter out statistical outliers such as homework assignments, we filter out the dataset entries according to the following rules:
* We filter out files which belong to the low-popular repos (the sum of stars and forks is less than 6)
* Next, we filter out files which belong to the repos with less than 5 Kotlin files
* Finally, we remove files which have less than 20 SLOC

We clean the content of the remaining dataset entries according to the following rules:
* We remove all non-ASCII entries
* We remove all package lines such as _package kotlinx.coroutines.channels_
* We remove half of the import lines.

# Model use

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load pre-trained model and tokenizer
model_name = 'JetBrains/CodeLlama-7B-KStack-full'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name).to('cuda')

# Create and encode input
input_text = """\
This function takes an integer n and returns factorial of a number:
fun factorial(n: Int): Int {\
"""
input_ids = tokenizer.encode(
    input_text, return_tensors='pt'
).to('cuda')

# Generate
output = model.generate(
    input_ids, max_length=60, num_return_sequences=1, 
    pad_token_id=tokenizer.eos_token_id,
)

# Decode output
generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
print(generated_text)
```

As with the base model, we can use FIM. To do this, the following format must be used: 
```
'<PRE> ' + prefix + ' <SUF> ' + suffix + ' <MID>'
```

# Training setup

The model was trained on one A100 GPU with following hyperparameters:

|         **Hyperparameter**           |             **Value**              |
|:---------------------------:|:----------------------------------------:|
|           `warmup`            |           5%            |
|        `max_lr`        |          1e-6          |
|        `num_epochs`        |          1          |
|        'attention_dropout' |          0.1        |
|        `scheduler`        |          cosine          |
|        `total_batch_size`        |          128 (~65K tokens per step)          |
|        `num_epochs`        |          1          |

More details about finetuning can be found in the technical report

# Fine-tuning data

For this model we used 15K exmaples of Kotlin Exercices dataset {TODO: link!}. Every example follows HumanEval like format. In total dataset contains about 3.5M tokens. 
For more information about the dataset follow the link.

# Evaluation 

To evaluate we used Kotlin Humaneval (more infromation here)

Fine-tuned model:

|         **Model name**           |             **Kotlin HumanEval Pass Rate**              |
|:---------------------------:|:----------------------------------------:|
|           `base model`            |           26.09            |
|        `fine-tuned model`        |          29.19         |

# Ethical Considerations and Limitations

Code Llama and its variants are a new technology that carries risks with use. The testing conducted to date could not cover all scenarios. For these reasons, as with all LLMs, Kexer's potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate or objectionable responses to user prompts. The model was fine-tuned on a specific data format (Kotlin tasks), and deviation from this format can also lead to inaccurate or undesirable responses to user queries. Therefore, before deploying any applications of Kexer, developers should perform safety testing and tuning tailored to their specific applications of the model.