--- license: mit datasets: - DKYoon/SlimPajama-6B - SWE-bench/SWE-smith-trajectories language: - en base_model: - Qwen/Qwen3-8B --- This are models trained based on [this paper](https://arxiv.org/abs/2307.06945). Pretrained on SlimPajama-6B and Fine-Tuned on SWE-smith trajectories by Claude 3.7 Sonnet. For more information see - code [here](https://github.com/JetBrains-Research/ICAE-for-SWE-agents) - Paper, descriptions and metircs are [here](https://github.com/Kirili4ik/implicit-context-compression-for-local-swe-agents-text/blob/main/build/master.pdf)