MagicQuant Releases Qwen3.8 27B Hybrid Quantized GGUF Models
crossivejoker · reddit · 2026-08-21
The author released Qwen3.8 27B GGUF models generated via the MagicQuant hybrid discovery system, utilizing Unsloth dynamic v3 and Imatrix. The post provides a detailed comparison between MagicQuant hybrid quantizations and original Unsloth versions in terms of KLD (divergence) and model size. Data shows that MagicQuant's hybrids (e.g., MQ-Q6K1) maintain extremely low KLD (0.000703) while optimizing volume. MagicQuant is a benchmark-driven GGUF evaluation and hybrid discovery system that finds high-performance hybrid quantization schemes by learning tensor configurations and running tests.
More from Infra
- Muse Spark Open runs locally on a Mac M3 Ultra via llama.cpp: frontier model on desktop — pcuenq · 2026-10-16
- Muse-Spark open-weights drop: 816B-A42B MoE model gets 359GB MLX 3-bit build — ns123abc · 2026-10-16
- vLLM ships Day-0 LoRA RL for Muse Spark Open, lifting AIME from 0.56 to 0.85 — vllm_project · 2026-10-16
- DSpark draft model for Muse Spark Open proposes 7 tokens per pass with 23-42% acceptance — vllm_project · 2026-10-16
- Cloudflare Stream's WebRTC ultra-low-latency live streaming is now Generally Available — ritakozlov · 2026-10-16
- YC-backed Overshoot launches inference optimization, claims $5M saved for AI teams — ycombinator · 2026-10-16