DeepSeek-V4-Flash-0731-CB-C16-NVFP4 (JasonW2025 NVFP4 build)

A third-party NVIDIA FP4-quantized redistribution of DeepSeek's open-weight DeepSeek-V4-Flash-0731 model, hosted by the Hugging Face user JasonW2025. The base model is a 284B-parameter Mixture-of-Experts LLM (13B active parameters) optimized for reasoning, coding, agentic workflows, and long-context chat up to approximately 1M tokens. Source: supabase.