DeepSeek-V-3.2

The 'DeepSeek-V-3.2' dataset/benchmark is used to evaluate the performance and trade-offs of different disaggregation strategies in large language model inference, particularly focusing on mixture-of-experts (MoE) models under various workload conditions.

Papers using DeepSeek-V-3.2 (2)

IS-CoT: Breaking the Long-form Generation Collapse via Interleaved Structural Thinking2026How Far Can Disaggregation Go? A Design-Space Exploration of Attention-FFN Disaggregation for Efficient MoE LLM Serving2026