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/stage_2_secure
    type: chat_template
    split: train
test_datasets:
  - path: felixwangg/stage_2_secure
    type: chat_template
    split: validation
dataset_prepared_path: /home/tkwang/scratch/SecSteer/axolotl-datasets/lora/Qwen2.5-Coder-7B/stage2-secure
val_set_size: 0
output_dir: /home/tkwang/scratch/SecSteer/axolotl-outputs/lora/Qwen2.5-Coder-7B-stage2-secure
sequence_len: 4096
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

adapter: lora
lora_model_dir: /home/tkwang/scratch/SecSteer/axolotl-outputs/lora/Qwen2.5-Coder-7B-stage1-combined/checkpoint-23
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-stage2-secure
wandb_log_model: "false"


gradient_accumulation_steps: 4
micro_batch_size: 4
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 2e-5

bf16: true
tf32: false

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:

plugins:

home/tkwang/scratch/SecSteer/axolotl-outputs/lora/Qwen2.5-Coder-7B-stage2-secure

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the felixwangg/stage_2_secure dataset. It achieves the following results on the evaluation set:

Loss: 0.7366
Ppl: 2.0889
Memory/max Active (gib): 38.25
Memory/max Allocated (gib): 38.25
Memory/device Reserved (gib): 56.53

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: 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: 4
training_steps: 46

Training results

Training Loss	Epoch	Step	Validation Loss	Ppl	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	0.7822	2.1862	37.91	37.91	41.87
0.7657	0.3315	15	0.7502	2.1174	38.25	38.25	55.38
0.7404	0.6630	30	0.7382	2.0921	38.25	38.25	56.53
0.769	0.9945	45	0.7366	2.0889	38.25	38.25	56.53

Framework versions

PEFT 0.18.1
Transformers 4.57.6
Pytorch 2.10.0+cu128
Datasets 4.5.0
Tokenizers 0.22.2

felixwangg/Qwen2.5-Coder-7B-stage2-secure

home/tkwang/scratch/SecSteer/axolotl-outputs/lora/Qwen2.5-Coder-7B-stage2-secure

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

felixwangg/
Qwen2.5-Coder-7B-stage2-secure