base_model: Qwen/Qwen3.5-9B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false

datasets:
  - path: felixwangg/glm-4.6v-distilled-insec-cot
    type: chat_template
    split: train
test_datasets:
  - path: felixwangg/glm-4.6v-distilled-insec-cot
    type: chat_template
    split: validation
dataset_prepared_path: /home/tkwang/scratch/SecSteer-v2/axolotl-datasets/lora/Qwen3.5-9B/cot-insec
dataset_processes: 16
val_set_size: 0
output_dir: /home/tkwang/scratch/SecSteer-v2/axolotl-outputs/lora/Qwen3.5-9B-cot-insec
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: cot-qwen3.5-primevul
wandb_entity: wtkuan
wandb_watch: "false"
wandb_name: Qwen3.5-9B-cot-insec
wandb_log_model: "false"

gradient_accumulation_steps: 8
micro_batch_size: 4
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 4e-05

bf16: true
tf32: false

train_on_inputs: false
roles_to_train: ['assistant']

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
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

ddp_find_unused_parameters: 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).
plugins:

home/tkwang/scratch/SecSteer-v2/axolotl-outputs/lora/Qwen3.5-9B-cot-insec

This model is a fine-tuned version of Qwen/Qwen3.5-9B on the felixwangg/glm-4.6v-distilled-insec-cot dataset. It achieves the following results on the evaluation set:

Loss: 0.8152
Ppl: 2.2597
Memory/max Active (gib): 56.35
Memory/max Allocated (gib): 56.35
Memory/device Reserved (gib): 75.5

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: 2
gradient_accumulation_steps: 8
total_train_batch_size: 64
total_eval_batch_size: 8
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: 56

Training results

Training Loss	Epoch	Step	Validation Loss	Ppl	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	0.9593	2.6099	55.99	55.99	60.75
0.8921	0.2715	15	0.8616	2.3668	56.35	56.35	75.37
0.7717	0.5430	30	0.8244	2.2804	56.35	56.35	75.49
0.8489	0.8145	45	0.8161	2.2617	56.35	56.35	75.5
0.8818	1.0	56	0.8152	2.2597	56.35	56.35	75.5

Framework versions

PEFT 0.19.1
Transformers 5.5.4
Pytorch 2.11.0+cu130
Datasets 4.5.0
Tokenizers 0.22.2

felixwangg/Qwen3.5-9B-cot-insec

home/tkwang/scratch/SecSteer-v2/axolotl-outputs/lora/Qwen3.5-9B-cot-insec

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

felixwangg/
Qwen3.5-9B-cot-insec