You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Segment-Credit GRPO: dense credit assignment from the policy's own confidence. +1.0 over standard GRPO at equal budget on medical exam reasoning, no teacher or reward model; built on a decontaminated SFT pipeline (Qwen2.5-7B, 78.8 → 84.3 on CMB-Exam), every gain paired-tested.
Own the full LLM lifecycle on an Apple M1 (8GB), no rented GPU: train a transformer from scratch, MLX-QLoRA + hand-rolled DPO on a <=1.5B model, quantize, serve behind a from-scratch micro-batcher, govern with eval + safety. Capstone 2/3.
Benchmark decontamination for LLM pretraining data: fast, auditable n-gram detection and removal of evaluation-benchmark text in training corpora (Rust).