hugpadhugpad

0xSero / DeepSeek-V4.1-Flash-Spark

Vision6367Updated Oct 3, 2026
Hugging Face

Downloads

636

Likes

7

Last updated

Oct 3, 2026

License

mit

DeepSeek-V4.1-Flash with every backbone routed expert re-encoded as EXL3 trellis (MUL1 codebook) at a per-expert K2 / K3 / K5 mix, sized to serve on two NVIDIA DGX Spark (GB10, 2 x 128 GB unified memory, tensor parallel 2). Average routed-expert rate 2.77 bpw. Everything else …

Tasks & library

  • image text to text
  • vllm

Tags

  • vllm
  • deepseek_v41
  • deepseek-v4.1
  • exl3
  • exllamav3
  • trellis

Technical specifications

ParametersNot available
Libraryvllm
Languagesen, zh
Base modeldeepseek-ai/DeepSeek-V4.1-Flash
Licensemit