PinnedMinyang Chen·Oct 2Put a big brain into a small model called QwengramIn my previous article, I reviewed how Strata handles the Qwen3.8-Flash-Next 125B model using a three-tier architecture that works together…
PinnedMinyang Chen·Oct 1Strata puts intelligence at scale on gaming PCs, using Qwen3.8-Flash-Next (125B-MoE)When you have a desktop PC with 32/64 GB of RAM and a consumer-grade GPU — such as a single or dual RTX 5060 Ti (16 GB), a single RTX 3090…A response icon2A response icon2
PinnedMinyang Chen·Aug 28A Deep Dive into DeepSeek Harness: Agent Loop, Plugins, Workflow, Ralph & Agent TeamDeepSeek Harness is an open-source AI agent runtime that provides the tools, memory, and environment LLMs need to carry out real-world…A response icon2A response icon2
PinnedMinyang Chen·Jul 19Research — Turning Market Noise into an Actionable ThesisStage 1 of the TradeVoiceLab Discovery Loop — Series
PinnedMinyang Chen·Jul 21Routine Work —Achieving Consistent Trading Results From Manual to AutomatedStage 6 of the TradeVoiceLab Discovery Loop — Series
Minyang Chen·6d agoStrata Drafting Model Local Evaluation — MTP vs. DFlash‑2Strata lets us run powerful AI locally on our own hardware.
Minyang Chen·Nov 1Experiment with AI-Native Animation and Story TellingA 38-Shot Animated Short on One Home GPU
Minyang Chen·Oct 26Unlocking the “Secret Speed Limit” of Older Nvidia GPUs (RTX 30 & 40 Series)One of the biggest news stories this week is the ability to unlock the hidden “speed limit” of older-generation Nvidia GPUs (specifically…
Minyang Chen·Sep 5NVIDIA PAIR — Why It Matters for Local AI Agents Running at Home with Multiple LLM ServersIf you have several LLM-capable machines at home or “personal AI cluster.” you’re currently facing several challenges:
Minyang Chen·Aug 26Qwen3.8–27B Models Optimized for Agentic WorksI’ve been extensively experimenting with Qwen3.8–27B since its release on local agentic workloads (including DeepSeek-Hardness), running it…A response icon2A response icon2
