fals3/peft-unit-test-generation-experiments

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.
url	A direct link to the model repository.
doi	The digital object identifier associated with the trained model.

Dataset Details

Dataset Description

Training Hyperparameters

Model-agnostic Hyperparameters

Hyperparameter	Method	Value
Common
Optimizer	-	AdamW
LR schedule	-	Linear
LR warmup ratio	-	0.1
Batch size	-	1
Gradient accumulation steps	-	8
# Epochs	-	3
Precision	-	Mixed
Learning rate	Full fine-tuning	5E-5
	LoRA	3E-4
	(IA)3	3E-4
	Prompt tuning	3E-3
Method specific
Alpha	LoRA	32
Dropout	LoRA	0.1
Rank	LoRA	16
Virtual tokens	Prompt tuning	20

Model-specific Hyperparameters

Hyperparameter	Method	Model	Value
Targeted attention modules	LoRA, (IA)3	codegen-350M-multi	qkv_proj
		Salesforce/codegen2-1B_P	qkv_proj
		Salesforce/codegen2-3_7B_P	qkv_proj
		Salesforce/codegen2-7B_P	qkv_proj
		Salesforce/codegen2-16B_P	qkv_proj
		meta-llama/CodeLlama-7b-hf	q_proj, v_proj
		bigcode/starcoderbase	c_attn
		bigcode/starcoder2-3b	q_proj, v_proj
		bigcode/starcoder2-7b	q_proj, v_proj
		bigcode/starcoder2-15b	q_proj, v_proj
Targeted feedforward modules	(IA)3	codegen-350M-multi	fc_out
		Salesforce/codegen2-1B_P	fc_out
		Salesforce/codegen2-3_7B_P	fc_out
		Salesforce/codegen2-7B_P	fc_out
		Salesforce/codegen2-16B_P	fc_out
		meta-llama/CodeLlama-7b-hf	down_proj
		bigcode/starcoderbase	mlp.c_proj
		bigcode/starcoder2-3b	q_proj, c_proj
		bigcode/starcoder2-7b	q_proj, c_proj
		bigcode/starcoder2-15b	q_proj, c_proj

Training Runs