nvidia-gb10

Reproducible refusal-subspace editing and TP=2 deployment of DeepSeek V4 Flash 0731 on two DGX Sparks

Native MiniMax-H3 video/audio inference in C — CUDA on Linux, Metal on Apple Silicon — with h3c studio, a self-hosted web UI.

Production-oriented Qwen3.6-35B-A3B-NVFP4-Fast vLLM deployment for NVIDIA DGX Spark / GB10

DeepSeek-V4-Flash + DSpark speculative decoding on a pair of NVIDIA DGX Sparks (vLLM TP=2 over RoCE) — tuned recipe, overlays that halve multi-turn TTFT, contamination-guarded benchmarks, ops runbook

One-command NVIDIA DGX Spark (GB10) cluster GPU monitoring with Grafana + Prometheus + DCGM + node_exporter + vLLM. Track GPU temperature, utilization, power, SM clock, memory, disk, network and LLM inference throughput in a pre-built dashboard. 一条命令搭建 DGX Spark 集群监控

Benchmark for GB10 - Nvidia DGX Spark

Measured Muse Glimmer 30B recipe for NVIDIA DGX Spark, with llama.cpp, DFlash parity checks, vision, tools, and GB10 benchmarks.

Reproducible high-speed Qwen3.6-35B-A3B inference on NVIDIA GB10

Auditable DeepSeek V4 Flash inference evidence on two NVIDIA GB10 systems

Three small, model-specific inference engines for NVIDIA DGX Spark.

Field notes, benchmarks, and turnkey scripts for running 284B LLM inference and multi-modal workflows across 2x NVIDIA DGX Spark (GB10) with 200GbE RoCE. 双机 NVIDIA DGX Spark (GB10) 284B 大模型分布式推理与多模态部署实战笔记、真机实测数据与避坑指南。

Reproducible Qwen3.8-27B Q4/Q6 and llama.cpp MTP benchmarks on NVIDIA GB10

Production-ready local AI agent stack for NVIDIA DGX Spark / ASUS GB10. Gemma 4 31B NVFP4 + bge-m3 embeddings + OpenClaw gateway, one docker compose up.

Weights-free two-DGX-Spark TP=2 SGLang recipe for Qwen3.8-Flash-Next-NVFP4

Headless MuJoCo on NVIDIA DGX Spark (GB10): EGL PNGs, Franka Panda, a cube. Documents why teleporting IK puts the cube through the hand. Collision-aware pinch: GRASP_FAIL. Verified August 2026.

Passive-first, read-only DGX Spark monitoring for unified memory, local models, thermals, queues, and private Tailnet access.

Atlas Inference Engine with Ling 3.0 Flash NVFP4 + MTP support for NVIDIA GB10 / DGX Spark

NVIDIA DGX Spark (GB10): systemd user profiles idle vs llm. Linger no longer reloads 100 GB of weights on every power button. Idle reboot proven. Verified August 2026.

Empirical NVIDIA GB10 and SM121 microarchitecture characterization

Auditable DeepSeek V4 Flash inference evidence on two NVIDIA GB10 systems