45 related articles
TutorialsComplete guide to deploying vLLM and SGLang locally. Compare performance vs LM Studio, deploy in 3 steps with Docker + AI assistant. Covers SGLang vs vLLM selection, 5090 VRAM optimization, and Cherry Studio integration.

In-depth feasibility analysis of deploying DeepSeek V4 Flash on two NVIDIA DGX Spark units offline, examining memory bandwidth, MoE communication overhead, and quantization strategies.

DeepSeek V4 Pro sparks open-source community buzz. Analysis of DeepSeek's V2-to-V3 evolution, MoE architecture cost advantages, and what developers should expect from the next-gen open-source LLM.

An insider's analysis of China's four AI labs — Qwen, DeepSeek, Moonshot, and Ling — revealing their distinct strategic bets on distribution, architecture, long-termism, and serving cost.

Deep analysis of AMD MI355X running Kimi K3 with superior cost-efficiency vs NVIDIA B300, and its implications for the AI inference hardware market.

Analyzing real LLM inference costs: from B200 GPU compute gains, vLLM framework optimization to MTP multi-token prediction, explaining why serving costs are widely overestimated.

Analysis of whether spending 20% more on hardware for self-hosting Kimi K3 to gain 20% task performance improvement is worthwhile, covering inference precision, VRAM optimization, and tiered deployment.

A deep dive into LLM inference cost structure and profitability models—from GPU throughput, MoE architecture, and KV Cache to scale effects—revealing the business logic behind API price wars.

Just $500 in RL fine-tuning enables a 9B open-source model to outperform frontier LLMs on catalog review tasks. Analysis of when small-model RL works and its enterprise implications.

Poolside releases Laguna S 2.1, an open-source agentic coding model: 118B total params with only 8B active, scoring 40.4% on DeepSWE — ~4.5x DeepSeek V4 Pro Max. Supports 1M-token context, deployable on a single workstation.

DeepSeek open-sources DeepSpec, a full-stack speculative decoding library delivering 60–85% inference speedup and up to 661% throughput gains in production. MIT license, 9 checkpoints, 9 benchmarks.

In-depth review of Panel AI v1.1.1: second-level installation, no-public-IP networking, batch compute cluster management. Learn how enterprise AI on-premises deployment barriers are dramatically lowered.

Deploy DeepSeek-V4-Flash DSpark on 8× H20-141G using GPUStack's SGLang backend on Day 0. Full walkthrough of Web UI config, parameter tuning, and 200 tokens/sec benchmark results.

QuantaMind is a free, open-source local AI Agent reliability testing tool using pass^k scoring and deterministic evaluation, supporting Ollama, llama.cpp, vLLM, and more.

From pressing Enter to the first character appearing, what happens inside an LLM? This article breaks down autoregressive generation, KV cache acceleration, and decoding strategies like temperature, Top-k, and Top-p.

SGLang officially integrates DSpark, solving the core pain point of speculative decoding failure under high-concurrency batches via confidence-driven variable-length verification. Supports Qwen3 and DeepSeek-V4, hitting 383.7 tok/s on B300.

The Miles team and AMD announce the full port of DeepSeek-V4 Flash RL training to AMD Instinct MI355X GPUs on ROCm, boosting AIME pass@1 from 0.39 to 0.49—a milestone for compute ecosystem diversity.

SGLang-Diffusion now officially supports LingBot-World 2.0, delivering leaps in resolution and temporal consistency. With live sessions, chunked streaming, and camera control, world models achieve low-latency controllable interaction.

An in-depth analysis of the vLLM inference framework's core principles: from the meaning of throughput (tokens/s), to the bottlenecks of autoregressive generation, to KV Cache, PagedAttention, and continuous batching.

MosiAI open-sources MOSS-Transcribe-Diarize-0.9B: unified speech transcription and speaker diarization, 128K context for 90-min audio, hotword boosting, SGLang Day-0 support, edge-deployable.