Kimi-K2 INT4MIX 模型 - Fastllm
Fastllm 的 Kimi-K2 INT4MIX 模型
https://github.com/ztxz16/fastllm
安装
pip install ftllm
下载模型：
ftllm download fastllm/Kimi-K2-Instruct-INT4MIX
运行模型
# 假设模型下载在 /root/Kimi-K2-Instruct-INT4MIX
ftllm run /root/Kimi-K2-Instruct-INT4MIX # 聊天模式
ftllm server /root/Kimi-K2-Instruct-INT4MIX # API 服务器模式（默认模型名称 = /root/Kimi-K2-Instruct-INT4MIX，端口 = 8080）
优化
单 CPU
如果您使用的是单个 CPU，请使用 -t 参数设置线程数（通常设置为 CPU 核心数 - 2）。
如果速度非常慢，可能是由于线程过多——考虑减少线程数。
例如：
ftllm server /root/Kimi-K2-Instruct-INT4MIX -t 12
多 CPU（多 NUMA 节点）
如果使用多路 CPU 的机器，您需要启用 CUDA + NUMA 异构加速模式。
使用环境变量 FASTLLM_NUMA_THREADS 设置线程数（通常设置为每个 NUMA 节点的核心数 - 2）。
如果性能非常慢，可能是由于线程过多——考虑减少线程数。
例如：
export FASTLLM_NUMA_THREADS=12 && ftllm server /root/Kimi-K2-Instruct-INT4MIX --device cuda --moe_device numa -t 1
English
Kimi-K2 INT4MIX model for Fastllm
https://github.com/ztxz16/fastllm
install
pip install ftllm
download model:
ftllm download fastllm/Kimi-K2-Instruct-INT4MIX
run model
# Assuming the model is downloaded in /root/Kimi-K2-Instruct-INT4MIX
ftllm run /root/Kimi-K2-Instruct-INT4MIX # chat
ftllm server /root/Kimi-K2-Instruct-INT4MIX # api server (default model_name = /root/Kimi-K2-Instruct-INT4MIX, port = 8080)
optimize
single CPU
If you are using a single CPU, set the number of threads with the -t parameter (generally set to CPU core count - 2).
If the speed is extremely slow, it may be due to too many threads—consider reducing them.
for example:
ftllm server /root/Kimi-K2-Instruct-INT4MIX -t 12
multi cpu (multi numa node)
If using a multi-socket CPU machine, you need to enable CUDA + NUMA heterogeneous acceleration mode.
Set the number of threads using the environment variable FASTLLM_NUMA_THREADS (typically set to the number of cores per NUMA node - 2).
If performance is extremely slow, it may be due to excessive threads—consider reducing them.
for example:
export FASTLLM_NUMA_THREADS=12 && ftllm server /root/Kimi-K2-Instruct-INT4MIX --device cuda --moe_device numa -t 1
- Downloads last month
- 14