llama-2-13b-hf-sensitivity-sorted

This model has sensitivity-based channel sorting applied. No quantization has been applied.

Configuration

Parameter	Value
Base model	meta-llama/Llama-2-13b-hf
Method	Sensitivity-based full channel sorting
Sorting order	Ascending (low sensitivity → high)
Quantization	None (sorting only)

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("jsyeom/sensitivity_sorted_models/llama-2-13b-hf-sensitivity-sorted")
tokenizer = AutoTokenizer.from_pretrained("jsyeom/sensitivity_sorted_models/llama-2-13b-hf-sensitivity-sorted")

Important Notes

• dst_ids.pt contains the full permutation indices for reference.
• sort_config.json contains metadata about the sorting configuration.

Reference

• QEFT: Quantization for Efficient Fine-Tuning of LLMs