hugpadhugpad

neopolita / Niwaki-Qwen3.8-Flash-Next-141B-A6B-v4.0-mlx

Text Generation73B1.3K2Updated Oct 15, 2026
Hugging Face

Downloads

1.3K

Likes

2

Last updated

Oct 15, 2026

License

other

Qwen3.8-Flash-Next at 141B total / ~5.9B active parameters, with its routed experts stored as lattice codes: 44.8 GB on disk, the same as v2.4, perplexity 10% below v2.4's, a task average of 0.794, the program-state behaviours we test at the unpruned model's level, and twenty …

Tasks & library

  • text generation
  • mlx

Tags

  • mlx
  • qwen4_exp
  • moe
  • pruning
  • quantization
  • mixture-of-experts

Technical specifications

Parameters73B
Librarymlx
LanguagesNot available
Base modelmlx-community/Qwen3.8-Flash-Next-4bit
Licenseother