How to Deploy vLLM on Kubernetes with GPUs: Step-by-Step (2026)
Deploy vLLM v0.32.0 on Kubernetes with NVIDIA GPUs: install GPU Operator v26.7.2, cache model weights on a PVC, serve an OpenAI-compatible API, autoscale on vllm:num_requests_waiting with KEDA, and fix CUDA out of memory and slow startup errors.
15 minOct 20, 2026How to Deploy an ML Model on Kubernetes with KServe 0.21 (2026)
Step-by-step KServe v0.21.0 guide: install in Standard mode with Helm, cert-manager and Gateway API, deploy scikit-learn and Hugging Face/vLLM InferenceServices, add HPA or KEDA autoscaling, canary rollouts, GPUs, and fix InferenceService not ready.
14 minOct 20, 2026AI Gateways in 2026: Six Options Compared After LiteLLM's 40-Minute Backdoor
What an AI gateway (LLM gateway) does in 2026 and how LiteLLM, Portkey, Kong, Cloudflare, Envoy AI Gateway (now Agent Router) and OpenRouter compare on routing, budgets, caching and guardrails, plus the LiteLLM PyPI backdoor, three LiteLLM CVEs on CISA KEV, and build vs buy.
13 minOct 5, 2026Model Collapse in 2026: The Loop Nobody Runs, a 40% Synthetic Phi-4, a 49.9% AI Web, and 300T Human Tokens Left
What the model-collapse research tested versus the headline: Shumailov's Nature loop deletes real data every generation and takes OPT-125m from 20 to 28 perplexity, while accumulating data bounds the error; my 100-generation Gaussian experiment shows median variance falling from 0.98 to 0.36 under replacement and holding at 0.98 under accumulation; how Phi-4 (40% synthetic), Llama 3, DeepSeek-R1 (800k samples), Qwen3, Kimi K2 and Nemotron-CC-v2 (38%) actually use it; Epoch's 300T-token stock, Graphite's 49.9% AI web, Cloudflare's 38,065:1 crawl ratio and the $1.5B Bartz settlement.
15 minOct 5, 2026MLOps in 2026: Evals Are the New Unit Tests, Judges Agree With Experts 66% of the Time, and OpenAI's Evals API Shuts Down on 30 November
The 2026 state of MLOps, checked against primary sources: OpenAI is closing its Evals API (read-only 31 October, off 30 November) and winding down fine-tuning while buying Promptfoo; Langfuse went to ClickHouse and W&B to CoreWeave; GDPval's model grader agrees with experts 66% of the time against 71% for humans; ARC-AGI-3 went from 0.51% to 99.9% in five months; Terminal-Bench retires a task when every frontier model solves it 5 of 5 times; MLflow 3.16, Kubeflow 26.03, Feast 0.66, lakeFS under BSL, and a pytest gate on a Wilson lower bound that I actually ran.
14 min