AITOPAI热报
ModelsOfficialSelected78°

DeepSeek V4.1-Flash model ships 1M context with extreme KV compression

DeepSeek-V4.1-Flash Ships Causal Encoder–Decoder MoE With 1M Context and Extreme KV Compression

官方一手pandaily@[email protected] (Pandaily)
Blog/Media
Read the original
Why we picked it

This matters for developers and researchers working with large-scale text data, as it significantly reduces memory requirements for long-form processing.

DeepSeek's V4.1-Flash model uses a 552B causal encoder-decoder MoE structure. It supports a 1M token context length and achieves extreme KV cache compression with CSA2+FP4 technology, reducing per-token memory usage to about 890 bytes.

Original · pandaily

DeepSeek-V4.1-Flash Ships Causal Encoder–Decoder MoE With 1M Context and Extreme KV Compression

DeepSeek’s V4.1-Flash pairs a 552B Causal Encoder–Decoder MoE with 1M context, CSA2+FP4 KV at ~890 bytes/token, and API access as deepseek-flash.

#DeepSeek#MoE#context length#AI models#LLM

Keep reading

Recent stories with the same tag or section
IndustryUS vs ChinaSelected10:25
DeepSeek Tokens Split Across 20 Inference Providers

Open-weight models decouple the model from the provider, so serving capacity becomes a commodity market where dozens of hosts compete on price and latency. Teams deploying DeepSeek can shop across providers rather than depend on one vendor. The 17% self-hosted share shows even the model's creator holds only a minority of its own serving traffic.

#DeepSeek#inference#open-source models
AI Will@FinanceYF5Source