Pinned Loading

1. CRISP CRISP Public

   [ACL'26 Findings] CRISP: Compressing Redundancy in Chain of Thought via Intrinsic Saliency Pruning

   Python 2

2. vocab_32k_gpt2_moe vocab_32k_gpt2_moe Public

   A from-scratch training stack for a GPT-2 style Mixture-of-Experts language model with a 32K SentencePiece vocabulary, covering the full alignment pipeline: pretraining, supervised fine-tuning, and…

   Python 5 1

3. vocab_32k_gpt2 vocab_32k_gpt2 Public

   A compact, end-to-end recipe for training a GPT-2 class language model from scratch — pretraining, supervised fine-tuning, and preference alignment.

   Python