News for hackers. Stuff that matters.

A comment on Kimi-K3 Releases on HuggingFace 7/27. Parent comment

Re:Kimi-K3 Releases on HuggingFace 7/27 by johndough on Monday July 27, 2026 @06:03AM (#49067331)
DeepSeek-V4 should use only 5GB for context due to CSA and HCA, see figure here: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro

But not every framework implements it properly yet.

[ Parent | on HN ]
Re:Kimi-K3 Releases on HuggingFace 7/27 by walrus01 on Monday July 27, 2026 @06:24AM (#49067496)
Yeah, or close enough to 5GB for estimation purposes, for example a just spawned qwen 3.5 122B llama-server instance reports as:

0.07.015.888 I common_memory_breakdown_print: | - Host | 170038 = 162913 + 6740 + 384 |

The 6740 is the cache size.

[ Parent | on HN ]