Qwen2.5-Coder-7B-Instruct-cpp-sec-step120-lam1

Combined Task Vector Model

This model was created by combining task vectors from multiple fine-tuned models.

Task Vector Computation

t_1 = TaskVector("Qwen/Qwen2.5-Coder-7B-Instruct", "/lustre10/scratch/tkwang/SecSteer/axolotl-outputs/lora-merged/plus-step-120")
t_2 = TaskVector("Qwen/Qwen2.5-Coder-7B-Instruct", "/lustre10/scratch/tkwang/SecSteer/axolotl-outputs/lora-merged/minus-step-120")
t_2 = TaskVector("Qwen/Qwen2.5-Coder-7B-Instruct", "None")
t_combined = 1.0 * t_1 + -1.0 * t_2
new_model = t_combined.apply_to("Qwen/Qwen2.5-Coder-7B-Instruct", scaling_coef=1.0)

Models Used

• Base Model: https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct
• Fine-tuned Model 1: https://huggingface.co//lustre10/scratch/tkwang/SecSteer/axolotl-outputs/lora-merged/plus-step-120
• Fine-tuned Model 2: https://huggingface.co//lustre10/scratch/tkwang/SecSteer/axolotl-outputs/lora-merged/minus-step-120
• Fine-tuned Model 3: https://huggingface.co/None

Technical Details

• Creation Script Git Hash: fb62f919e9796b294f1ffb6297b05d11fa945ac0
• Task Vector Method: Additive combination
• Args: { "pretrained_model": "Qwen/Qwen2.5-Coder-7B-Instruct", "finetuned_model1": "/lustre10/scratch/tkwang/SecSteer/axolotl-outputs/lora-merged/plus-step-120", "finetuned_model2": "/lustre10/scratch/tkwang/SecSteer/axolotl-outputs/lora-merged/minus-step-120", "finetuned_model3": null, "apply_to_diff_model_architecure": null, "output_model_name": "felixwangg/Qwen2.5-Coder-7B-Instruct-cpp-sec-step120-lam1", "output_dir": "/lustre10/scratch/tkwang/SecSteer/axolotl-outputs/weight-arithmetic/step-120/lambda-1", "scaling_coef": 1.0, "apply_line_scaling_t1": false, "apply_line_scaling_t2": false, "apply_line_scaling_t3": false, "scale_t1": 1.0, "scale_t2": -1.0, "scale_t3": null }