ClaudioSavelli/FAME-topics_KLM_llama32-3b-instruct-qa

TEXT GENERATIONConcurrency Cost:1Model Size:3.2BQuant:BF16Ctx Length:32kPublished:Apr 2, 2026License:otherArchitecture:Transformer Cold

ClaudioSavelli/FAME-topics_KLM_llama32-3b-instruct-qa is a 3.2 billion parameter instruction-tuned language model, based on the Llama-3.2-3B-Instruct architecture. This model is specifically unlearned using the KL Minimization method for the FAME-topics setting. It is designed for question-answering tasks within this specialized context, offering a focused approach to information retrieval and response generation.

Loading preview...

Model Overview

ClaudioSavelli/FAME-topics_KLM_llama32-3b-instruct-qa is a 3.2 billion parameter instruction-tuned model derived from the meta-llama/Llama-3.2-3B-Instruct base. Its primary distinction lies in its application of the KL Minimization method for "unlearning" within the FAME-topics setting. This process aims to modify the model's knowledge or behavior in a targeted way, making it particularly relevant for specific research or application areas where such unlearning techniques are beneficial.

Key Characteristics

  • Base Model: Built upon the robust Llama-3.2-3B-Instruct architecture.
  • Parameter Count: Features 3.2 billion parameters, balancing performance with computational efficiency.
  • Context Length: Supports a substantial context window of 32768 tokens, allowing for processing longer inputs and maintaining conversational coherence.
  • Specialized Training: Utilizes KL Minimization for unlearning, as detailed in the associated research paper.

Intended Use Cases

This model is particularly suited for:

  • Research in Model Unlearning: Ideal for exploring and evaluating the effectiveness of KL Minimization in modifying LLM behavior.
  • FAME-topics Applications: Designed for question-answering within the specific FAME-topics domain, where its unlearning process is directly relevant.
  • Specialized QA Systems: Can be integrated into systems requiring a model with a modified knowledge base or specific response characteristics due to unlearning.