Library Breakdowns
Open-source ML libraries. Dissected for deep understanding.
TrainingPython
rLLM
Framework for post-training language agents via RL — a Workflow abstraction that lets you train arbitrary multi-turn agentic programs with GRPO, powering models like DeepCoder (60.6% LiveCodeBench) and DeepSWE (59% SWEBench-Verified).
reinforcement-learningagentspost-training+5
InferencePython
LMCache
KV cache management layer for LLM serving engines — 3-10x TTFT reduction through tiered storage, prefix sharing, and non-prefix cache reuse.
inferencekv-cachevllm+3