Skip to content
BestHub
Sign in
Tagged articles

MoE Models

6 articles · Page 1 of 1
Lao Guo's Learning Space
Lao Guo's Learning Space
Nov 11, 2026 · Artificial Intelligence

Unsloth Dynamic v3.1 Quantization Fits 552B MoE on Mac: Memory Requirements by Bit Level

Unsloth's Dynamic v3.1 quantization reduces Qwen3.8-Flash-Next to 68GB (1-bit) and DeepSeek-V4.1-Flash-1017 to 128-145GB (1-bit), enabling them to run on Macs with 96GB+ unified memory, while benchmarks show 50% fewer tokens, 7% faster speed, and 15% higher accuracy versus same-size quantizations.

DeepSeek-V4.1-Flash-1017Dynamic QuantizationLLM Deployment
0 likes · 11 min read
Unsloth Dynamic v3.1 Quantization Fits 552B MoE on Mac: Memory Requirements by Bit Level
Lao Guo's Learning Space
Lao Guo's Learning Space
Oct 31, 2026 · Artificial Intelligence

GLM-5.3 Flash vs DeepSeek-V4.1 Flash: 20% Faster Code Gen on M5 Ultra

Benchmark comparison of GLM-5.3 Flash and DeepSeek-V4.1 Flash on Mac Studio M5 Ultra 256GB shows GLM-5.3 Flash achieves 84.9 vs 71.2 tok/s in code decoding (1.2x), 109.2 vs 93.6 tok/s in 8-request concurrency, and mixed first-token latency results across 16K-128K contexts, with caveats about different inference engines and no quality evaluation.

DeepSeek V4.1 FlashGLM-5.3 FlashMac Studio M5 Ultra
0 likes · 8 min read
GLM-5.3 Flash vs DeepSeek-V4.1 Flash: 20% Faster Code Gen on M5 Ultra
Machine Heart
Machine Heart
Oct 23, 2026 · Artificial Intelligence

TRACE: Rollout-Guided FP4 Quantization Makes MoE RL Training Fast and Stable, 5.4× Speedup

Alibaba Qwen and Ohio State University propose TRACE, a rollout-guided quantization-aware training method that aligns training-side FP4 rounding with actual rollout quantization results, enabling stable FP4 W/A and KV cache rollout for MoE models with up to 5.4× decoding throughput gains and performance matching BF16 baselines.

Alibaba QwenFP4 quantizationMoE Models
0 likes · 9 min read
TRACE: Rollout-Guided FP4 Quantization Makes MoE RL Training Fast and Stable, 5.4× Speedup
Lao Guo's Learning Space
Lao Guo's Learning Space
Sep 23, 2026 · Artificial Intelligence

Mac mini 16GB to Studio 256GB: Which Mac Runs Your LLMs Best?

This article analyzes Mac mini and Mac Studio configurations for local LLM inference, showing how unified memory capacity determines which models fit and memory bandwidth dictates token generation speed, with real-world benchmarks for 8B to 235B models across M6, M5 Pro, M5 Max, and M5 Ultra chips.

Apple SiliconLLM inferenceMac Studio
0 likes · 13 min read
Mac mini 16GB to Studio 256GB: Which Mac Runs Your LLMs Best?
Architects' Tech Alliance
Architects' Tech Alliance
Apr 30, 2026 · Artificial Intelligence

Token Era Unpacked: The ‘One Chip, Two Models, Three Clouds’ Blueprint for AI Agents

The article analyzes how the rise of AI agents transforms the industry from dialogue‑centric models to 24/7 digital employees, driving a shift toward CPU‑centric compute, domestic MoE models with strong coding abilities, and cloud platforms that become the core deployment and billing ecosystem, all fueled by massive token inflation.

AI AgentsAI hardwareMoE Models
0 likes · 13 min read
Token Era Unpacked: The ‘One Chip, Two Models, Three Clouds’ Blueprint for AI Agents