Indexed by TopGit from live GitHub metadata: ztxz16/fastllm has 4.9k stars, written primarily in C++. fastllm是后端无依赖的高性能大模型推理库。同时支持张量并行推理稠密模型和混合模式推理MOE模型,任意10G以上显卡即可推理满血DeepSeek。双路9004/9005服务器+单显卡部署DeepSeek满血满精度原版模型,单并发20tps;INT4量化模型单并发30tps,多并发可达60+。
Snapshot summary built from the project's own GitHub metadata — there's no written TopGit review yet. The page will update automatically when a full review is published.
WHY NO REVIEW YET
TopGit writes full reviews for the most-starred, most-requested repositories. This page is a snapshot until then — see the READ ME tab for the original README in full.
说明: 一般和device指定为不同的设备实现混合推理,例如
--device cuda --moe_device cpu来实现MOE模型的单卡+CPU混合推理。
--device cuda --moe_device numa 来实现MOE模型的单卡+多NUMA节点加速推理
如果指定的模型不是moe结构的模型,这个参数不会生效
export FASTLLM_NUMA_THREADS=27 && ftllm server fastllm/DeepSeek-V3-0324-INT4 --device cuda --moe_device numa -t 1
# 使用多numa推理,每个numa节点使用27个线程
export FASTLLM_NUMA_THREADS=16 && ftllm server fastllm/DeepSeek-V3-0324-INT4 --device cuda --moe_device numa -t 1
# 使用多numa推理,每个numa节点使用16个线程
numactl -C 0-31 -m 0 ftllm server fastllm/DeepSeek-V3-0324-INT4 --device cuda --moe_device cpu -t 27
# 绑定单numa节点,使用CPU推理,使用27线程
不同硬件上,不同参数发挥出的性能有很大不同。一般而言,CPU上使用的线程数不建议超过物理核数
5. 其它参数
--moe_experts:
描述: 指定 MOE(Mixture of Experts)层使用的专家数。不设定则根据模型配置设定。减少专家数可以提高推理速度,但可能降低推理准确度
示例: --moe_experts 6
--cuda_se:
描述: 指定 MOE中的共享专家 是否在cuda上执行,默认为true
示例: --cuda_se false
--cuda_slab:
描述: 设置 CUDA 模型权重 slab 分配块大小,单位 MB。默认 0 表示关闭。将大量 MOE 专家权重放在 CUDA 上时,可用它减少小权重分别 cudaMalloc 带来的显存页对齐和碎片开销。
No homepage URL was recorded for ztxz16/fastllm in TopGit's last sync. The README tab above frequently contains screenshots and demo links, or check the repository description on GitHub.
Does ztxz16/fastllm have any tags?
TopGit's last sync did not record any GitHub topics for ztxz16/fastllm. GitHub topics appear in the right sidebar of a repository page; that's the authoritative place to check.
How active is development on ztxz16/fastllm?
The most recent commit recorded on ztxz16/fastllm was 12 days ago, based on the GitHub push timestamp. The repository has 486 forks — one of the better signals of community interest.
How many stars does ztxz16/fastllm have?
ztxz16/fastllm has 4.9k GitHub stars — refresh the page for the live number, or check github.com/ztxz16/fastllm. TopGit mirrors GitHub's count but does not claim minute-by-minute accuracy.
What language is ztxz16/fastllm written in?
ztxz16/fastllm is written primarily in C++. GitHub's language field is based on the largest share of bytes in the default branch.
What license does ztxz16/fastllm use?
ztxz16/fastllm is released under the Apache-2.0 license. Always verify the LICENSE file directly on GitHub for the authoritative terms — license strings can be edited out of sync with a project's actual stance.
Where do I read more about ztxz16/fastllm?
This TopGit page is a snapshot — the READ ME tab shows the project's own README content (links stripped, images preserved). The GitHub repository at github.com/ztxz16/fastllm is the definitive source.
Read full README in the tab above.
Want a second opinion on fastllm?
Ask an AI that can read this page — one click and you get its take on fastllm.