Zhanyl Abdybaeva https://zhanyl-tech.github.io/ Recent content on Zhanyl Abdybaeva Hugo en-us Sun, 26 Jul 2026 00:00:00 +0000 Epilog GPU Validator https://zhanyl-tech.github.io/projects/epilog-gpu-validator/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/epilog-gpu-validator/ Drain a node for a persistently faulty GPU — and never for a transient one. Agentic ML Inference Infrastructure https://zhanyl-tech.github.io/projects/agentic-ml-inference-infrastructure/ Thu, 01 Jan 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/agentic-ml-inference-infrastructure/ Production agentic AI infrastructure layer: autonomous systems that reason, decide, and act on real workloads using vLLM, LangGraph, and MCP. Slinky GitOps https://zhanyl-tech.github.io/projects/slinky-gitops/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/slinky-gitops/ Slurm on Kubernetes from nothing in one command — and the auth-key rotation nobody wants to test in production. 128-GPU Distributed Training Platform https://zhanyl-tech.github.io/projects/distributed-training-platform/ Thu, 01 Jan 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/distributed-training-platform/ Distributed ML training infrastructure for large-scale models: FSDP, gradient checkpointing, mixed precision, and cost-optimized multi-node orchestration. IB Slurm Exporter https://zhanyl-tech.github.io/projects/ib-slurm-exporter/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/ib-slurm-exporter/ Correlate InfiniBand and RoCE counters with the Slurm job that owns them, so a slow collective can be traced to the fabric. GPU-Accelerated Volatility Surface Calibration https://zhanyl-tech.github.io/projects/gpu-volatility-surface/ Thu, 01 Jan 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/gpu-volatility-surface/ CUDA-accelerated local and stochastic volatility surface calibration — Dupire's model, Heston, and hybrid approaches for real-time derivatives pricing. GPU Reaper https://zhanyl-tech.github.io/projects/gpu-reaper/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/gpu-reaper/ Find wasted GPU allocations on a Slurm cluster and escalate them through alert, drain, and cancel. Slurm Scheduler Lab https://zhanyl-tech.github.io/projects/slurm-scheduler-lab/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/projects/slurm-scheduler-lab/ Test Slurm priority and backfill policy against a job trace before it reaches a live controller. I Was Planning a Slurm 25.11 Upgrade. Then 26.05 Shipped. https://zhanyl-tech.github.io/experiments/2026-07-26-slurm-25-11-upgrade-notes/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/experiments/2026-07-26-slurm-25-11-upgrade-notes/ Three versions, two config renames that break on restart, and one cgroup change that quietly breaks every job-attribution tool you own. Your Slurm Priority Weights Matter Less Than Your Users' Time Limits https://zhanyl-tech.github.io/experiments/2026-07-26-slurm-backfill-time-limits/ Sun, 26 Jul 2026 00:00:00 +0000 https://zhanyl-tech.github.io/experiments/2026-07-26-slurm-backfill-time-limits/ I built a simulator to tune Slurm priority weights. It kept telling me the weights were not the problem. LServe and SampleAttention: What Sparse Attention Actually Changes in Prefill and Decode https://zhanyl-tech.github.io/blog/2026-01-30-lserve-sampleattention-sparse-attention/ Fri, 30 Jan 2026 00:00:00 +0000 https://zhanyl-tech.github.io/blog/2026-01-30-lserve-sampleattention-sparse-attention/ Structured sparsity aligned with GPU attention kernels: LServe’s unified serving stack for prefill and decode, and SampleAttention’s empirical patterns plus CRA as a runtime quality floor. vLLM vs TensorRT-LLM: First Throughput Numbers on Llama-3 70B https://zhanyl-tech.github.io/experiments/2026-01-20-vllm-vs-tensorrt-llm-first-look/ Tue, 20 Jan 2026 00:00:00 +0000 https://zhanyl-tech.github.io/experiments/2026-01-20-vllm-vs-tensorrt-llm-first-look/ Quick benchmark comparing vLLM and TensorRT-LLM throughput on a single A100 80GB with Llama-3 70B. Setup notes, raw numbers, and what I'd investigate next. How KV-Cache Paging Works in vLLM — and Why It Matters for Production https://zhanyl-tech.github.io/blog/2026-01-15-kv-cache-paging-vllm/ Thu, 15 Jan 2026 00:00:00 +0000 https://zhanyl-tech.github.io/blog/2026-01-15-kv-cache-paging-vllm/ A technical walkthrough of PagedAttention: the memory management innovation that makes vLLM practical for serving many concurrent LLM sessions without fragmenting GPU memory. About https://zhanyl-tech.github.io/about/ Mon, 01 Jan 0001 00:00:00 +0000 https://zhanyl-tech.github.io/about/ About me Reading https://zhanyl-tech.github.io/reading/ Mon, 01 Jan 0001 00:00:00 +0000 https://zhanyl-tech.github.io/reading/ Books I recommend