I'm trying to post-train AliceAI-Foundation-80B-A3B-Base myself

A developer is attempting to post-train the AliceAI-Foundation-80B-A3B-Base model using three 32GB V100 GPUs. Their method involves shallow distillation of Qwen 3.8 27b, followed by Supervised Fine-Tuning (SFT) with synthetic data generated for long-horizon agent tasks. Reinforcement Learning (RL) via GRPO and an evaluation model is planned next. The developer is considering using a more powerful model for training data generation, but aims to keep the entire process local.