See axolotl config

axolotl version: 0.13.2

base_model: Qwen/Qwen2.5-Coder-7B-Instruct
model_type: Qwen2ForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false

datasets:
  - path: felixwangg/prime_vul_plus_splitted
    type: chat_template
    split: train
test_datasets:
  - path: felixwangg/prime_vul_plus_splitted
    type: chat_template
    split: validation
dataset_prepared_path: ~/SecSteer/axolotl-datasets/lora/Qwen2.5-Coder-7B/sft-plus-alpha-0
val_set_size: 0
output_dir: ~/SecSteer/axolotl-outputs/lora/Qwen2.5-Coder-7B-sft-plus-alpha-0-lr4e-5
sequence_len: 4096
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 16
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
merge_lora: true

wandb_project: sft-primevul-sweep-ctx-0
wandb_entity: wtkuan
wandb_watch: "false"
wandb_name: Qwen2.5-Coder-7B-sft-plus-alpha-0
wandb_log_model: "false"

gradient_accumulation_steps: 4
micro_batch_size: 4
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 4e-5

bf16: true
tf32: false

# roles_to_train: ['user', 'system', 'assistant']

train_on_inputs: false
roles_to_train: ['assistant']

gradient_checkpointing: true
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

num_epochs: 1
warmup_ratio: 0.1
early_stopping_patience: 1000
eval_steps: 15
save_steps: 15
save_total_limit: 1000
load_best_model_at_end: true
# 
weight_decay: 0.02
special_tokens:

# SecCodeBench C/CPP benchmark evaluation after every validation step.
# Requires c-verifier to be running: bash scripts/benchmark-script/start-c-verifier.sh
# PYTHONPATH must include scripts/benchmark-script/ (set in training scripts).
# Note: axolotl uses 'plugins:' (not 'callbacks:') to load custom TrainerCallbacks.
plugins:

~/SecSteer/axolotl-outputs/lora/Qwen2.5-Coder-7B-sft-plus-alpha-0-lr4e-5

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the felixwangg/prime_vul_plus_splitted dataset. It achieves the following results on the evaluation set:

• Loss: 0.7204
• Ppl: 2.0554
• Memory/max Active (gib): 38.24
• Memory/max Allocated (gib): 38.24
• Memory/device Reserved (gib): 51.91

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: 4e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 5
• training_steps: 57

Training results

Training Loss	Epoch	Step	Validation Loss	Ppl	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	0.8204	2.2715	37.91	37.91	41.87
0.7468	0.2655	15	0.7747	2.1698	38.24	38.24	51.33
0.6478	0.5310	30	0.7270	2.0689	38.24	38.24	51.91
0.777	0.7965	45	0.7204	2.0554	38.24	38.24	51.91

Framework versions

• Transformers 4.57.6
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2