paper retrieval bm25 rag llm-agent meta zero-shot

paper deepseek moe llm-architecture attention training inference

paper alignment safety midtraining synthetic-data generalization

IR retrieval benchmark LLM reasoning oblique latent

ai workflow video codex gpt-image product-demo

deepseek parallel-strategy moe compute-communication-overlap distributed-training engineering

llm architecture training transformer

architecture ssm agent-memory transformer-alternative inference-optimization