---
dataset_info:
  features:
  - name: model_type
    dtype: string
  - name: namespace
    dtype: string
  - name: model_name
    dtype: string
  - name: training_method
    dtype: string
  - name: model_size
    dtype: int64
  - name: trainable_params
    dtype: int64
  - name: url
    dtype: string
  - name: doi
    dtype: float64
  splits:
  - name: train
    num_bytes: 6257
    num_examples: 40
  download_size: 4879
  dataset_size: 6257
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
pretty_name: PEFT Unit Test Generation Experiments
size_categories:
- n<1K
---
# PEFT Unit Test Generation Experiments

## Dataset description

The **PEFT Unit Test Generation Experiments** dataset contains metadata and details about a set of trained models used for generating unit tests with parameter-efficient fine-tuning (PEFT) methods. This dataset includes models from multiple namespaces and various sizes, trained with different tuning methods to provide a comprehensive resource for unit test generation research.

## Dataset Structure

### Data Fields

Each example in the dataset corresponds to a specific trained model variant and includes the following features:

| Feature Name      | Description                                                                                         |
|-------------------|-----------------------------------------------------------------------------------------------------|
| `model_type`      | The type or architecture of the base model (e.g., codegen, starcoder).                              |
| `namespace`       | The organization or group that created or published the base model (e.g., Salesforce, meta-llama). |
| `model_name`      | The specific name or identifier of the model.                                                      |
| `training_method` | The parameter-efficient fine-tuning method used for training (e.g., full fine-tuning, LoRA, IA³).  |
| `model_size`      | The size of the model, typically measured in number of parameters (e.g., 350M, 7B).                 |
| `trainable_params`| The number of trainable parameters for the specific tuning method and [hyperparameters](#training-hyperparameters). |
| `url`             | A direct link to the model repository.                                           |
| `doi`             | The digital object identifier associated with the trained model.                   |

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

### Training Hyperparameters

#### Model-agnostic Hyperparameters
<table>
  <thead>
    <tr>
      <th>Hyperparameter</th>
      <th>Method</th>
      <th>Value</th>
    </tr>
  </thead>
  <tbody>
    <tr style="font-weight: bold;">
      <td colspan="3">Common</td>
    </tr>
    <tr>
      <td>Optimizer</td>
      <td>-</td>
      <td>AdamW</td>
    </tr>
    <tr>
      <td>LR schedule</td>
      <td>-</td>
      <td>Linear</td>
    </tr>
    <tr>
      <td>LR warmup ratio</td>
      <td>-</td>
      <td>0.1</td>
    </tr>
    <tr>
      <td>Batch size</td>
      <td>-</td>
      <td>1</td>
    </tr>
    <tr>
      <td>Gradient accumulation steps</td>
      <td>-</td>
      <td>8</td>
    </tr>
    <tr>
      <td># Epochs</td>
      <td>-</td>
      <td>3</td>
    </tr>
    <tr>
      <td>Precision</td>
      <td>-</td>
      <td>Mixed</td>
    </tr>
    <tr>
      <td style="vertical-align: middle;" rowspan="4">Learning rate</td>
      <td>Full fine-tuning</td>
      <td>5E-5</td>
    </tr>
    <tr>
      <td>LoRA</td>
      <td>3E-4</td>
    </tr>
    <tr>
      <td>(IA)<sup>3</sup></td>
      <td>3E-4</td>
    </tr>
    <tr>
      <td>Prompt tuning</td>
      <td>3E-3</td>
    </tr>
    <tr style="font-weight: bold;">
      <td colspan="3">Method specific</td>
    </tr>
    <tr>
      <td>Alpha</td>
      <td>LoRA</td>
      <td>32</td>
    </tr>
    <tr>
      <td>Dropout</td>
      <td>LoRA</td>
      <td>0.1</td>
    </tr>
    <tr>
      <td>Rank</td>
      <td>LoRA</td>
      <td>16</td>
    </tr>
    <tr>
      <td>Virtual tokens</td>
      <td>Prompt tuning</td>
      <td>20</td>
    </tr>
  </tbody>
</table>

#### Model-specific Hyperparameters
<table>
  <thead>
    <tr>
      <th>Hyperparameter</th>
      <th>Method</th>
      <th>Model</th>
      <th>Value</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td rowspan="10" style="vertical-align: middle;">Targeted attention modules</td>
      <td rowspan="10" style="vertical-align: middle;">LoRA, (IA)<sup>3</sup></td>
      <td>codegen-350M-multi</td>
      <td>qkv_proj</td>
    </tr>
    <tr><td>Salesforce/codegen2-1B_P</td><td>qkv_proj</td></tr>
    <tr><td>Salesforce/codegen2-3_7B_P</td><td>qkv_proj</td></tr>
    <tr><td>Salesforce/codegen2-7B_P</td><td>qkv_proj</td></tr>
    <tr><td>Salesforce/codegen2-16B_P</td><td>qkv_proj</td></tr>
    <tr><td>meta-llama/CodeLlama-7b-hf</td><td>q_proj, v_proj</td></tr>
    <tr><td>bigcode/starcoderbase</td><td>c_attn</td></tr>
    <tr><td>bigcode/starcoder2-3b</td><td>q_proj, v_proj</td></tr>
    <tr><td>bigcode/starcoder2-7b</td><td>q_proj, v_proj</td></tr>
    <tr><td>bigcode/starcoder2-15b</td><td>q_proj, v_proj</td></tr>
    <tr>
      <td rowspan="10" style="vertical-align: middle;">Targeted feedforward modules</td>
      <td rowspan="10" style="vertical-align: middle;">(IA)<sup>3</sup></td>
      <td>codegen-350M-multi</td>
      <td>fc_out</td>
    </tr>
    <tr><td>Salesforce/codegen2-1B_P</td><td>fc_out</td></tr>
    <tr><td>Salesforce/codegen2-3_7B_P</td><td>fc_out</td></tr>
    <tr><td>Salesforce/codegen2-7B_P</td><td>fc_out</td></tr>
    <tr><td>Salesforce/codegen2-16B_P</td><td>fc_out</td></tr>
    <tr><td>meta-llama/CodeLlama-7b-hf</td><td>down_proj</td></tr>
    <tr><td>bigcode/starcoderbase</td><td>mlp.c_proj</td></tr>
    <tr><td>bigcode/starcoder2-3b</td><td>q_proj, c_proj</td></tr>
    <tr><td>bigcode/starcoder2-7b</td><td>q_proj, c_proj</td></tr>
    <tr><td>bigcode/starcoder2-15b</td><td>q_proj, c_proj</td></tr>
  </tbody>
</table>


## Training Runs

![image/png](https://huggingface.co/datasets/fals3/peft-unit-test-generation-experiments/resolve/main/assets/full-fine-tuning-train-loss-log.png)

![image/png](https://huggingface.co/datasets/fals3/peft-unit-test-generation-experiments/resolve/main/assets/lora-train-loss-log.png)

![image/png](https://huggingface.co/datasets/fals3/peft-unit-test-generation-experiments/resolve/main/assets/ia3-train-loss-log.png)

![image/png](https://huggingface.co/datasets/fals3/peft-unit-test-generation-experiments/resolve/main/assets/prompt-tuning-train-loss-log.png)