PreciseCoder/PreciseCoder-27B-GRPO-0.5F1

Partial fine-tune vs Qwen/Qwen3.6-27B. Partial change: 89% of the weights changed (attention, mlp, norms), in 64 of 64 layers.

Tensors changed: Embeddings 0/4, MLP 192/304, Attention 357/556, lm_head 0/1, Norms 125/319

2 derivatives checked