27-method post-training spectrum, distilled to 5 production-grade artifacts + 1 live demo. Every number is a receipt — negative results included.
gogogo
jjshin
·
AI & ML interests
LLM post-training & MLOps in production · 27-method fine-tuning spectrum (LoRA · ReFT · KTO · GRPO · RLCD) · alert-triage decision systems that ship · 三平台作品位: ModelScope theGAI / GitHub jjxyxsjr
Recent Activity
published a model 22 days ago
jjshin/kev-0.8B-LoRA-ranker updated a model 22 days ago
jjshin/kev-0.8B-LoRA-ranker liked a model 22 days ago
jjshin/laya-fullft-typed-decisionsOrganizations
None yet
models 7
jjshin/kev-0.8B-LoRA-ranker
Text Classification • Updated • 6
jjshin/Qwen3.5-4B-GRPO-MultiTurn-MATH-fix
Text Generation • Updated
jjshin/laya-fullft-typed-decisions
Text Classification • 0.4B • Updated • 1
jjshin/kev-9B-LoRA-ranker
Text Ranking • Updated • 35
jjshin/Qwen3.5-4B-ReFT-gsm8k
Text Generation • Updated • 23
jjshin/Qwen3-8B-KTO-ultrafeedback
Text Generation • Updated • 23
jjshin/jjshin
Updated
datasets 0
None public yet