Okapi zh-BLOOM-1.7B JudgeRM LoRA

A LoRA adapter checkpoint fine-tuned atop the BLOOM-1.7B multilingual base model under the Okapi instruction-tuning and RLHF framework, targeting Chinese-language alignment. It represents a lightweight task-specific adapter (trained for 500 steps at sequence length 1024) rather than a full retrained model. Source: supabase.