ggml/llama.cpp releases: b11870

TL;DR

The ggml/llama.cpp project released a new model update incorporating DeepSeek-V4.1-Flash which optimizes memory usage and improves computational efficiency, emphasizing the importance of enhanced model performance in large language models.

Detailed Summary

The ggml/llama.cpp project released a new model version incorporating DeepSeek-V4.1-Flash, which removes MXFP8 and converts the Engram table to BF16, optimizing memory usage. This update also pre-computes compressed-KV read layers for caching and utilizes get_cells() for shared-history checks, enhancing efficiency without relying on llama. The broader impact includes improved performance and reduced computational load in language modeling applications.

Key Points

• • model : add DeepSeek-V4.1-Flash (deepseek41)
• • deepseek41 : remove MXFP8, convert the Engram table to BF16
• • deepseek41 : compute the compressed-KV read layers once in the cache
• • deepseek41 : use get_cells() for the shared-history check, drop llama