Хирургия весов Qwen2.5-0.5B → гибрид 3:1 KDA/GQA: CPT, recall-серии и экономика длинного контекста (память ×3.9, decode ×3.65 на 512K, ≤32K).
Roman Nekrasov
Rob1234567
AI & ML interests
Areas of interest: agentic mid-training, reinforcement learning with reward verification (RLVR), scaling agent environments, interleaved agent reasoning with tools
Recent Activity
updated a dataset 5 days ago
Rob1234567/kda-graft-econ-evidence-2026-09-12 updated a collection 5 days ago
kda-graft: KDA-графт Qwen2.5-0.5B updated a collection 5 days ago
kda-graft: KDA-графт Qwen2.5-0.5BOrganizations
None yet