Roman Nekrasov

Rob1234567

romannekrasovaillm

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 Space 4 days ago

Rob1234567/personal-manifesto

published a Space 4 days ago

Rob1234567/personal-manifesto

updated a collection 9 days ago

Axiom

View all activity

Organizations

None yet

updated a Space 4 days ago

Personal Manifesto of an AI Researcher in the Agentic Era

🧭

Roman Nekrasov · architect-researcher

published a Space 4 days ago

Personal Manifesto of an AI Researcher in the Agentic Era

🧭

Roman Nekrasov · architect-researcher

updated a collection 9 days ago

Axiom

Collection

Axiom JAX/MXFP4 pilot — now with the MFU-55 research card and its trace dataset: 452 → 1,203.7 tok/s on one DGX Spark in 8 days. • 3 items • Updated 9 days ago

updated a Space 9 days ago

The Kernel War for MFU

⚡

28 agent cycles, one JAX LLM, one DGX Spark

updated a dataset 9 days ago

Rob1234567/axiom-mfu55-traces

Updated 9 days ago • 59

published a Space 9 days ago

The Kernel War for MFU

⚡

28 agent cycles, one JAX LLM, one DGX Spark

published a dataset 9 days ago

Rob1234567/axiom-mfu55-traces

Updated 9 days ago • 59

updated a model 23 days ago

Rob1234567/axiom-pilot-l3full-step300

Updated 23 days ago

published a model 23 days ago

Rob1234567/axiom-pilot-l3full-step300

Updated 23 days ago

updated a collection 23 days ago

Axiom

Collection

Axiom JAX/MXFP4 pilot — now with the MFU-55 research card and its trace dataset: 452 → 1,203.7 tok/s on one DGX Spark in 8 days. • 3 items • Updated 9 days ago

updated a dataset about 2 months ago

Rob1234567/kda-graft-econ-evidence-2026-09-12

Updated Sep 13 • 1.19k

updated a collection about 2 months ago

kda-graft: KDA-графт Qwen2.5-0.5B

Collection

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). • 2 items • Updated Sep 12

updated a model about 2 months ago

Rob1234567/kda-graft-qwen2.5-0.5b-linear-attention-3to1

Text Generation • 0.5B • Updated Sep 12 • 90

published a model about 2 months ago

Rob1234567/kda-graft-qwen2.5-0.5b-linear-attention-3to1

Text Generation • 0.5B • Updated Sep 12 • 90

published a dataset about 2 months ago

Rob1234567/kda-graft-econ-evidence-2026-09-12

Updated Sep 13 • 1.19k

updated a collection 3 months ago

Qwen3.5-4B Ariadna — доменный ML/AI-эксперт

Collection

Qwen3.5-4B, дообучен на 124K концептов arXiv. safetensors (GRPO v10) + GGUF Q4_K_M с PDF-отчётами и кейсами. • 2 items • Updated Aug 1

updated a model 3 months ago

Rob1234567/qwen3.5-4b-ariadna-grpo-v1-gguf

Text Generation • 4B • Updated Aug 1 • 62 • 1

Roman Nekrasov

AI & ML interests

Recent Activity

Organizations

Rob1234567's activity

Personal Manifesto of an AI Researcher in the Agentic Era

Personal Manifesto of an AI Researcher in the Agentic Era

The Kernel War for MFU

The Kernel War for MFU