Jeesup/svd-safety-l2_remove50_disc_b005

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kPublished:Sep 10, 2026License:llama2Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Jeesup/svd-safety-l2_remove50_disc_b005 is a Llama-2-7b-chat checkpoint compressed using SVD-LLM, retaining 50.5% of its original parameters. This 7 billion parameter model then had 0.5% of its parameters restored using the 'disc' component selection rule. It serves as a research artifact to study how SVD compression impacts safety behavior and the effectiveness of different component-selection rules in recovery, rather than being a general-purpose chat model.

Loading preview...

Model Overview

Jeesup/svd-safety-l2_remove50_disc_b005 is a specialized research artifact derived from meta-llama/Llama-2-7b-chat-hf. This model has undergone significant compression using the SVD-LLM method, reducing its parameter count to 50.5% of the original 7 billion parameters. Following compression, a small budget of 0.5% of the dense parameters was used to restore SVD components, specifically selected by the disc rule.

Key Characteristics

  • Base Model: Llama-2-7b-chat-hf
  • Compression Method: SVD-LLM, removing 49.52% of parameters.
  • Restoration: 0.5% parameter budget restored using the disc selection rule, involving 3044 components.
  • Resulting Parameter Fraction: 50.48% of the original dense parameters.

Intended Use and Limitations

This model is not intended for general-purpose chat applications. Its primary purpose is as an experimental subject within a research study to measure safety and utility trade-offs under model compression. The study aims to quantify how compression degrades safety (e.g., increased attack-success rates) and to test various recovery mechanisms. As such, this specific checkpoint may exhibit deliberately degraded safety behavior compared to the original Llama-2-7b-chat. Users should treat it as an experimental subject and conduct their own evaluations before drawing conclusions.