Unsloth AI releases NVFP4 quantized Qwen3.6 models, offering up to a 2.5x speedup inside 24GB of VRAM · Digg

/TECH

Unsloth AI releases NVFP4 quantized Qwen3.6 models, offering up to a 2.5x speedup inside 24GB of VRAM

C🤗

The optimized models reach 17,561 tokens per second on Blackwell GPUs.

Combined views

14K

3 posts, first seen 9h ago

103 likes

Unsloth AI releases NVFP4 quantized Qwen3.6 models, offering up to a 2.5x speedup inside 24GB of VRAM

C🤗

Commentary on X

Highest ranked

Mozart Louis@madebymozartTECH

@UnslothAI @Alibaba_Qwen Honest unsloth quants are the best I’ve used so far

Combined views

14K

3 posts, first seen 9h ago

103 likes

5 comments

45 saves

195 reposts

Sentiment

Positive100%0%Negative

Positive Read

Many users expressed gratitude and excitement toward Unsloth and Alibaba_Qwen for releasing NVFP4 quants on Qwen3.6 models because they deliver substantial 2.5x GPU speedups and strong throughput results.

Based on 36 visible X reactions from 151 accounts.

Useful Links

NVFP4 - a unsloth Collection(huggingface)viamerve
Qwen3.6 - How to Run Locally | Unsloth Documentation(unsloth.ai)viamerve

Digg Deeper

Ask a question below.

Published answers will appear here.

Commentary on X

Highest ranked

Mozart Louis@madebymozartTECH

@UnslothAI @Alibaba_Qwen Honest unsloth quants are the best I’ve used so far

shaun@BiggestTECH

@UnslothAI @Alibaba_Qwen Extremely impressive! Great to see more improvements

Neo@NeoAIForecastTECH

@UnslothAI @Alibaba_Qwen Unsloth, unreal! Thanks team

Taeyun Jang@taeyun16_TECH

@UnslothAI @Alibaba_Qwen 항상 오픈소스 모델 성능 향상에 노력해주셔서 감사합니다!

View all

Daniel Han@danielhanchenTECH

2.5x faster NVFP4 Qwen3.6 27B + 35B-A3B quants! We applied our Unsloth dynamic method to NVFP4 using W4A4 vs W4A16. Also 2 35B-A3B versions - NVFP4 (1.56x faster) and NVFP4-Fast (1.79x faster). Fast uses pure W4A4, whilst non fast uses a mixture to stay a bit more accurate. https://twitter.com/UnslothAI/status/2075566124687892597