OLAResearchX/xmemtransfer-mistral-from-llama2-r4-25m25m

XMemTransfer Mistral adaptor from LLaMA-2 source (R4, 25M/25M)

This repository contains a target-side adaptor checkpoint for the XMemTransfer cross-model QA setting.

Project links

• Website: XMemTransfer project page
• Paper: Cross-Model Memory Transfer via Target-Side Reader Adaptation
• Code: OLAResearch/XMemTransfer
• Model repo: https://huggingface.co/OLAResearchX/xmemtransfer-mistral-from-llama2-r4-25m25m
• Collection: https://huggingface.co/collections/OLAResearchX/xmemtransfer

Files

• adaptor_best.pt: best target-side adaptor checkpoint
• config.json: adaptor configuration
• results.json: summary metrics from the run

Related source memory

This adaptor is loaded on top of mistralai/Mistral-7B-v0.3 and uses the frozen source memory released at OLAResearchX/xmemtransfer-llama2-source-25m.

Provenance

• source memory: LLaMA-2 7B WikiText-2021 source memory
• target model family: Mistral 7B
• reader setting: dual-layer, 4-branch
• token budgets: 25M source / 25M target-side fitting

Usage and limitations

These files are research artifacts for the XMemTransfer implementation, not standalone Transformers pipeline checkpoints. Follow the project code to construct the memory reader and load the checkpoint. Results depend on the paired backbone, source memory, data preprocessing, and configuration released with the model.

Citation

Please cite the XMemTransfer paper when using these artifacts.