 Claude
|
dd77232b16
|
Add 12 AI infrastructure and LLM operations skills
New skills covering hot-topic AI engineering subjects:
Local AI Infrastructure:
- vllm-server: High-throughput LLM serving with PagedAttention, tensor parallelism, quantization
- llm-inference-scaling: KEDA-based GPU autoscaling for LLM inference on Kubernetes
- rag-infrastructure: Production RAG with hybrid search, reranking, and embedding pipelines
- llm-fine-tuning: QLoRA/LoRA fine-tuning with Axolotl, DeepSpeed ZeRO-3, and DPO alignment
Infrastructure:
- gpu-server-management: NVIDIA driver setup, MIG partitioning, DCGM monitoring
- vector-database-ops: Qdrant, Weaviate, pgvector for production AI search
- llm-gateway: LiteLLM-based API gateway with rate limiting, virtual keys, fallback routing
DevOps/AI:
- llm-cost-optimization: Model right-sizing, prompt/semantic caching, batch API, break-even analysis
- llm-caching: Multi-layer exact + semantic + provider caching to cut costs 30-70%
- ai-pipeline-orchestration: Prefect/Airflow/Dagster for RAG ingestion and training workflows
Orchestration:
- model-serving-kubernetes: KServe + Triton with canary deployments and GPU autoscaling
Security:
- ai-security-hardening: Prompt injection defense, PII scrubbing, model weight verification
https://claude.ai/code/session_011MN1C4PrkCeg2Qmi7q1ZUe
|
2026-03-02 01:18:29 +00:00 |
|