Blog
Writing on GPUs, LLMs, MLOps, Kubernetes — and mindset · 3517 posts
#2026-03 765#english 592#culture 264#deep-dive 254#kubernetes 247#career 229#ai 216#llm 208#devops 193#2026-04 146#security 141#database 114#observability 113#communication 109#history 107#architecture 100#productivity 96#finance 88#economy 84#mindset 81#psychology 80#ai-papers 79#food 78#it 78#travel 78#deep-learning 77#japanese 77#networking 77#performance 72#business-travel 70#linux 70#gpu 69#ai-agent 66#cs-fundamentals 63#postgresql 60#rag 58#self-improvement 55#learning 53#mlops 53#ai-platform 51
FDE Interview Prep — Diagnosis Scenarios, Customer Simulations, Case Studies
A Forward Deployed Engineer (FDE) interview often replaces algorithm puzzles with incident-diagnosis scenarios on an unfamiliar system, and culture-fit questions with an angry-customer role play — because what it measure
2026-08-12 · 5 min read #career#fde#forward-deployed-engineer#interview#job-searchInside vLLM (5) — Prefix Caching, and Why System Prompt Design Is Performance
A rundown of when vLLM prefix caching hits and when it misses, based on the official design document. Covers how block hashes chain to the block before them, why only full blocks get cached, how a single timestamp placed
2026-08-12 · 7 min read #vllm#prefix-caching#kv-cache#prompt-engineering#llmInside vLLM (3) — How Continuous Batching Keeps the GPU Busy
Explains why static batching leaves the GPU idle, and how vLLM iteration-level scheduling removes that idle time. Covers the difference in character between prefill and decode, V1 unified scheduler and its token budget,
2026-08-12 · 7 min read #vllm#continuous-batching#chunked-prefill#llm#inferenceInside vLLM (7) — Deployment Tuning, Common Pitfalls, and OOM Triage
A practical order for tuning a vLLM deployment. Covers what gpumemoryutilization actually sets, when to use tensor parallelism versus pipeline parallelism, how to choose quantization and a KV cache data type, and a diagn
2026-08-12 · 10 min read #vllm#gpu#quantization#tensor-parallel#llmInside vLLM (2) — Why PagedAttention Splits the KV Cache Into Pages
Explains why PagedAttention splits the KV cache into fixed-size blocks, verified against the original paper (arXiv:2309.06180) and the official vLLM design documentation. Covers the internal and external fragmentation th
2026-08-12 · 7 min read #vllm#paged-attention#kv-cache#llm#gpuInside vLLM (6) — Context Window vs max_model_len vs max_tokens, Fully Explained
If the difference between context window and maxtokens has been confusing, this single post sorts it out. Covers what each of the following actually limits, with a comparison table and real error messages: the context wi
2026-08-12 · 12 min read #vllm#context-window#max-model-len#max-tokens#llmInside vLLM (4) — The Scheduler and Preemption, Where Throughput Collapses
Explains what the vLLM scheduler decides on every step, how the waiting queue and running list move, and how preemption, triggered when the KV cache runs short, wrecks latency and throughput, verified against the officia
2026-08-12 · 7 min read #vllm#scheduler#preemption#kv-cache#llmInside vLLM (1) — The Full Path From One Request to One Token
Follows the full path a single request takes through vLLM, from arrival to the first token out. Covers what the API server, scheduler, KV cache manager, worker, and sampler each do, and how the process layout splits afte
2026-08-12 · 6 min read #vllm#llm#inference#gpu#ai-platformVideo Generation and Understanding Technical Reports: What to Read, and Why Constraints Beat Demos
Ten video generation and understanding technical reports, each verified by opening the arXiv abstract page directly. CogVideoX, Movie Gen, HunyuanVideo, LTX-Video, Wan, Seedance 1.0 and 2.0, plus Qwen2.5-VL, VideoLLaMA 3
2026-08-12 · 7 min read #ai-papers#paper-review#technical-report#video-generation#diffusion-transformerText LLM Technical Reports: What to Read, and How to Read Design Decisions Instead of Rankings
Nine text-LLM technical reports, each verified by opening the arXiv abstract page directly. What DeepSeek-V3, DeepSeek-R1, Qwen3, Gemma 3, Olmo 3, Kimi K2, MiniMax-01, Mellum2 and s1 each did that was new, and which limi
2026-08-12 · 7 min read #ai-papers#paper-review#technical-report#llm#moeThe Training Recipe — From Pre-training to Post-training, and What Reports Write Down
Comparing, exactly as written in the reports, the three stages and six context extensions of Llama 3, the three-stage pre-training of Qwen3, the learning-rate schedule and two-phase YaRN extension of DeepSeek-V3, and the
2026-08-12 · 6 min read #ai-papers#model-internals#pretraining#training-recipe#data-mixturePositional Encoding — RoPE and the Price of Context Extension
Showing through wavelength arithmetic how a single ropetheta value acts on context length, then comparing the 500000 of Llama 3, the ABF step in Qwen3, the YaRN settings of DeepSeek-V3 and Kimi K2, and the partial rotati
2026-08-12 · 6 min read #ai-papers#model-internals#rope#positional-encoding#long-contextAnatomy of config.json — Reading a Model From One Settings File
How every field in config.json — hiddensize, numhiddenlayers, numattentionheads versus numkeyvalueheads, headdim, intermediatesize, ropetheta, vocabsize, tiewordembeddings — shows up in memory and speed, and a hand count
2026-08-12 · 6 min read #ai-papers#model-internals#config-json#transformer#llm-architectureOCR and Document Understanding Technical Reports: What to Read, and Why Parsing Is Not Finished
Ten OCR and document understanding technical reports, each verified by opening the arXiv abstract page directly. From Donut and Nougat through GOT-OCR2.0, olmOCR, DeepSeek-OCR and its successor, GLM-OCR, Qianfan-OCR, Min
2026-08-12 · 7 min read #ai-papers#paper-review#technical-report#ocr#document-aiHow to Read Leaderboards and Benchmarks: Why SOTA Has Such a Short Shelf Life
Twelve benchmark methodology papers, each verified by opening the arXiv abstract page directly, assembled into a guide for reading leaderboard numbers. Data contamination, prompt format sensitivity, eval harness differen
2026-08-12 · 7 min read #ai-papers#paper-review#technical-report#benchmark#evaluationReading a Technical Report Critically — What Gets Written Down and What Goes Missing
How to separate the verifiable from the unverifiable in a model technical report. Covers the limits of self-reported benchmarks, the places where a config and a report disagree, values that could not be read from gated r
2026-08-12 · 7 min read #ai-papers#model-internals#tech-report#benchmarks#evaluationMoE Routing — How an Expert Gets Picked
Reading the config fields of a mixture-of-experts layer against real models. Comparing Mixtral 2-of-8, Qwen3 8-of-128, the 256 routed experts plus a shared expert in DeepSeek-V3, and the sparsity of 48 in Kimi K2, then w
2026-08-12 · 6 min read #ai-papers#model-internals#mixture-of-experts#moe#routingSpeech Recognition and Synthesis Technical Reports: What to Read, and What a Single WER Hides
Ten speech recognition and synthesis technical reports, each verified by opening the arXiv abstract page directly. Whisper, Omnilingual ASR, Qwen3-ASR, the Open ASR Leaderboard, Seed-TTS and F5-TTS, CosyVoice 2, Qwen3-TT
2026-08-12 · 7 min read #ai-papers#paper-review#technical-report#speech-recognition#text-to-speechImage Generation and Understanding Technical Reports: What to Read, and Why Design Beats Sample Images
Eleven image generation and understanding technical reports, each verified by opening the arXiv abstract page directly. Rectified flow transformers, VAR, Emu3, SANA, Janus-Pro, FLUX.1 Kontext, the Qwen-Image line, Seedre
2026-08-12 · 8 min read #ai-papers#paper-review#technical-report#image-generation#diffusion-transformerInside the Tokenizer — Why Korean Costs More Tokens, and What That Costs
How byte-level BPE works, then downloading the actual tokenizer files of Qwen3, DeepSeek-V3, and Mixtral to tokenize the same English and Korean text and compare. Covers the vocabulary-size tradeoff, why the config vocab
2026-08-12 · 6 min read #ai-papers#model-internals#tokenizer#bpe#korean-nlp