ttt-e2e-long-context

【开发与工程】Enable long-context modeling via test-time training with meta-learning. Inner loop continues training on context, compressing information into weights rather than KV cache, outer loop optimizes initialization—maintaining full-attention quality with RNN-like constant inference latency across 8K-128K token contexts.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准