-
## Background
The [WiscKey](https://www.usenix.org/system/files/conference/fast16/fast16-papers-lu.pdf) paper showed the value of segregating storage for large values in order to reduce write ampli…
-
Any thoughts/plans about YaRN support for the positional embeddings?
https://github.com/jquesnelle/yarn
I don't actually see them beat regular linear scaling w/ fine-tuning in the paper, but presu…
-
### Idea
Use int4 as the compression technique to fit larger models onto Navi machines or possibly MI series machines. Weights would be compressed using encoding scheme that would pack two 4 bits n…
-
```
Endpoint#write *1
$4
info
Redis#write $2688
# Storage info
sophia.version = 1.2.3
sophia.build = 59ff278
sophia.error =
sophia.path = ./serenity_db
sophia.path_create = 1
memory.limit = 107374182…
-
Greetings,
I am running SSDB on my server, the data directory is about 1.9GB, all the items are simple KV records , the length of keys and values are about 10B and 20B
I found the the ssdb occupy a l…
-
你好,8核24G内存ssd的机器运行单个ssdb,配置如何优化? compression设置为true后内存(cache_size + 10 \* write_buffer_size \* 66 + 32)超过物理内存会出现什么现象,谢谢
-
Hi, thanks for your insightful work! But it seems that some settings and implementations of baselines are not so proper and fair?
1. Window size. In `run_longbench.py: 221`, PyramidKV uses window s…
-
rootid->supervoxels
this can be cached essentially permanently since rootids mutate whenever the mapping changes
however they are quite large, so the object store could quickly start taking a l…
-
This is actually not my question, llama.cpp wants to implement this but encountered some problems.
https://github.com/ggerganov/llama.cpp/pull/4207
-
### Is your proposal related to a problem?
Yes. Figuring out which metrics are exported by the various Thanos components and the meaning behind them is often a process that requires diving deep…