🚀 DeepSeek Unveils New Model 'MODEL1' on Anniversary of DeepSeek-R1
#DeepSeek #MODEL1 #DeepSeekR1 #FlashMLA #GitHub #V32 #DeepSeekV32 #KVCacheLayout #SparsityHandling #FP8Decoding #MemoryManagement
On January 21, DeepSeek marked the first anniversary of DeepSeek-R1 by revealing details about its new model, 'MODEL1.' According to BlockBeats, the company updated its FlashMLA code on GitHub, highlighting 28 mentions of MODEL1 across 114 files. This model appears alongside V32, indicating it is distinct from DeepSeek-V3.2. The differences in the code are evident in areas such as KV cache layout, sparsity handling, and FP8 decoding, with several optimizations in memory management.#DeepSeek #MODEL1 #DeepSeekR1 #FlashMLA #GitHub #V32 #DeepSeekV32 #KVCacheLayout #SparsityHandling #FP8Decoding #MemoryManagement