MedSwin/MedSwin-Merged-TA-SFT-0.7

This is a merge of pre-trained language models created using mergekit.

• Developed by: Medical Swinburne University of Technology AI Team
• Funded by: Swinburne University of Technology
• Language(s): English
• License: Apache 2.0

Merge Details

Merge Method

This model was merged using the Task Arithmetic merge method using medalpaca-7b as a base.

Task Arithmetic: Merging LLMs with Vector Algebra

Task Arithmetic is a powerful model merging technique where the knowledge learned during fine-tuning on a specific task is isolated as a task vector.

• Task Vector Calculation: This vector represents the change in weights from the base model to the fine-tuned model.

  Task Vector = Weights_Fine-Tuned - Weights_Base

• Merging: To combine multiple skills, it add the task vectors for each skill back to the base model's weights.

  Model_Merged = Weights_Base + sum(Task Vectors)

• Editing/Forgetting: To remove a specific learned capability, you subtract its task vector from the model's weights.

---

This technique enables simple, linear arithmetic operations in the model's high-dimensional parameter space to efficiently compose, edit, or remove specific capabilities without retraining.

---

Would you like to know which popular model merging framework implements the Task Arithmetic technique?

Models Merged

The following models were included in the merge:

• MedSwin-7B-KD

• MedSwin-7B-SFT

• Benchmark Dataset

• Benchmark Logs

Configuration

The following YAML configuration was used to produce this model:

base_model: medalpaca-7b
dtype: bfloat16
merge_method: task_arithmetic
modules:
  default:
    slices:
    - sources:
      - layer_range: [0, 32]
        model: medalpaca-sft
        parameters:
          weight: 0.7
      - layer_range: [0, 32]
        model: medalpaca-kd
        parameters:
          weight: 0.7
      - layer_range: [0, 32]
        model: medalpaca-7b

Review all model metrics benchmark via Benchmark Document Preview.