See axolotl config

axolotl version: 0.8.0.dev0

base_model: bigcode/starcoder2-15b
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: kevinwsbr/vulnfixes-web
    type: alpaca


dataset_prepared_path:
val_set_size: 0.05
output_dir: ./outputs/starcoder-vulnfixes-web

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: starcoder
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 3
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 2e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16: false
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 20
evals_per_epoch: 4
eval_steps:
eval_table_size:
saves_per_epoch: 4
save_steps:
save_total_limit: 2
debug:
deepspeed:
weight_decay:
fsdp:
fsdp_config:
special_tokens:
  pad_token: "<|endoftext|>"
  eos_token: "<|endoftext|>"

outputs/starcoder-vulnfixes-web

This model is a fine-tuned version of bigcode/starcoder2-15b on the kevinwsbr/vulnfixes-web dataset. It achieves the following results on the evaluation set:

• Loss: 0.0529

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 20
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.1499	0.0092	1	0.0645
0.1554	0.2569	28	0.0622
0.0745	0.5138	56	0.0571
0.0616	0.7706	84	0.0559
0.0645	1.0275	112	0.0547
0.0601	1.2844	140	0.0542
0.0688	1.5413	168	0.0537
0.0424	1.7982	196	0.0534
0.086	2.0550	224	0.0532
0.0759	2.3119	252	0.0530
0.0583	2.5688	280	0.0529
0.1087	2.8257	308	0.0529

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0