DeepSeek V3

Headlines · 2

1. Nov 3
   Trade-offs in Distributed Inference for Attention and MoE

   The report analyzes distributed-inference strategies across tensor, data, expert, context, and pipeline parallelism using weight, KV-cache, activation, communication, capacity, and latency budgets. Its author argues that GQA can shard KV heads up to the KV-head count, while MLA's shared latent cache must be replicated under tensor parallelism, and compares MoE tensor parallelism with expert parallelism in terms of communication, memory use, load balancing, and implementation complexity. The report title labels it installment 2, while the body calls it the third of eight installments in an inference-infrastructure methodology series. Its comparative figures and conclusions are the author's analysis and are not independently corroborated by the supplied evidence.

   掘金 · 人工智能 · 🔥 0
2. Oct 9
   DiffuSpace raises hundreds of millions of yuan in record dLLM funding round

   Shenzhen diffusion-language-model startup DiffuSpace has completed two consecutive funding rounds co-led by Matrix Partners China, Shunwei Capital and Legend Capital, with CAS Star, Huawei Hubble and Horizon Robotics among the followers. The company did not disclose the exact amount; reports describe it as "hundreds of millions of yuan," while TMTPost says the total is close to RMB 500 million and calls it a record for global diffusion language model (dLLM) funding. Founded in Shenzhen in May 2026 by HKU NLP lab co-director Kong Lingpeng and his PhD students Gong Sansan and Ye Jiacheng, DiffuSpace says its earlier Dream 7B model has passed 2.5 million Hugging Face downloads and that a larger next-generation dLLM is being trained for release and open-sourcing soon.

   36氪 人工智能 · 🔥 0

Experience and discussion from the community

Share my DeepSeek V3 experienceAsk about DeepSeek V3