Yz

qwen3-next

The first installment in the Qwen3-Next series with strong performance in terms of both parameter efficiency and inference speed.

Araç Kullanımı Akıl Yürütme 80b
Hızlı Kurulum (Ollama kuruluysa) 
ollama run qwen3-next

Ollama kurulu değil mi? ollama.com/download — Windows, macOS ve Linux için ücretsiz. İlk çalıştırmada model indirilir, sonrası tamamen çevrimdışıdır.

Varyantlar

Boyut büyüdükçe kalite artar, donanım ihtiyacı yükselir. Başlangıç için küçük varyantı deneyin.

EtiketBoyutBağlamGirdiKomut
latest 50GB 256K Text ollama run qwen3-next:latest
80b 50GB 256K Text ollama run qwen3-next:80b
80b-a3b-instruct-q4_K_M 50GB 256K Text ollama run qwen3-next:80b-a3b-instruct-q4_K_M
80b-a3b-instruct-q8_0 85GB 256K Text ollama run qwen3-next:80b-a3b-instruct-q8_0
80b-a3b-instruct-fp16 159GB 256K Text ollama run qwen3-next:80b-a3b-instruct-fp16
80b-a3b-thinking 50GB 256K Text ollama run qwen3-next:80b-a3b-thinking
80b-a3b-thinking-q4_K_M 50GB 256K Text ollama run qwen3-next:80b-a3b-thinking-q4_K_M
80b-a3b-thinking-q8_0 85GB 256K Text ollama run qwen3-next:80b-a3b-thinking-q8_0
80b-a3b-thinking-fp16 159GB 256K Text ollama run qwen3-next:80b-a3b-thinking-fp16

Model Detayları ve Benchmarklar (kaynak: ollama.com)

image.png

Qwen3-Next-80B-A3B is the first installment in the Qwen3-Next series and features the following key enhancements:

  • Hybrid Attention: Replaces standard attention with the combination of Gated DeltaNet and Gated Attention, enabling efficient context modeling for ultra-long context length.
  • High-Sparsity Mixture-of-Experts (MoE): Achieves an extreme low activation ratio in MoE layers, drastically reducing FLOPs per token while preserving model capacity.
  • Stability Optimizations: Includes techniques such as zero-centered and weight-decayed layernorm, and other stabilizing enhancements for robust pre-training and post-training.
  • Multi-Token Prediction (MTP): Boosts pretraining model performance and accelerates inference.