Qwen3-30B-A3B — Serving Quantization - a ThakiCloud Collection

ThakiCloud 's Collections

Thaki-HW — Hardware Design LLM (Research Preview)

EmbeddingGemma 2 Edge — Quantization

RAG-Gate — Answer / Retrieve / Stop Gates for RAG

ChainCheck — Chain-Sensitive Evidence Judges

PACE — Enterprise Policy Decisions: Learned or Reasoned

Merchant Desk — Korean Small-Business Customer Desk

Qwen-Image-2.1 — Few-Step Distillation

K-Decision — Korean typed-decision adapters (weights only)

K-MT — Korean-Specialized Translation Without Forgetting

Satoori-KO — Korean Regional Dialect 27B

Evals & Reproducibility

Language Confusion Suppression — Vocabulary Pruning

Vision-Language — Serving Quantization

Qwen3-30B-A3B — Serving Quantization

Qwen3-Coder-30B-A3B — Serving Quantization

Qwen3.8-27B — Serving Quantization

SkillRet — Agent Skill Retrieval

Human-KO — Korean Style-Aligned 27B

Qwen3-30B-A3B — Serving Quantization

updated Sep 10

General-purpose MoE 30B-A3B quantization. Three builds differing only in algorithm — also a rebuild-noise comparison.

ThakiCloud/Qwen3-30B-A3B-NVFP4

Text Generation • 16B • Updated Sep 7 • 40

Note NVFP4 baseline
ThakiCloud/Qwen3-30B-A3B-NVFP4-GPTQ

Text Generation • 31B • Updated Sep 7 • 57

Note NVFP4 with GPTQ calibration
ThakiCloud/Qwen3-30B-A3B-NVFP4-RTN

Text Generation • 31B • Updated Sep 7 • 88

Note NVFP4 with RTN (no calibration, control arm)