收起左侧

AIMAX395平台NPU使用的BUG修复

0
回复
16
查看
[ 复制链接 ]

0

主题

0

回帖

0

牛值

江湖小虾

提交问题前,请尽量一次性提供完整信息。
信息不完整、无明确复现路径、缺少关键日志的帖子,将不予处理并直接关闭;请补充完整信息后重新发帖。
请先按下方格式填写;除以下基础信息外,也请根据你遇到的问题补充截图、录屏、报错提示、复现步骤等信息,便于更快定位和处理。
应用中心第三方开发者上架应用问题可以联系对应开发者反馈。

系统版本:X86

设备环境:物理机、局域网、系统V1.2.0201

BUG现象:硬件是AMD的AIMAX395平台,安装了商店的ROCM7.2驱动和AI相关的程序。但是系统自带的NPU固件缺失。

出现频率:必现,我通过自编译的AMD官方NPU驱动修复后,只要更新系统还会出现。

联系方式:13301688161

日志文件:2026-07-10 17:35:51.424 [Info] (FastFlowLM) Found flm at: flm
2026-07-10 17:35:51.445 [Error] (Router) Backend load failed: FLM NPU validation failed: NPU firmware is incompatible. Kernel version is incompatible.
2026-07-10 17:35:51.445 [Error] (Router) Visit https://lemonade-server.ai/driver_install.html for driver installation instructions.
2026-07-10 17:35:51.445 [Warn] (Router) Load failed with non-file-not-found error, evicting all models and retrying...
2026-07-10 17:35:51.445 [Info] (Router) Evicting all models (0 total)
2026-07-10 17:35:51.445 [Info] (Router) Evict all completed. Remaining tombstoned models: 0
2026-07-10 17:35:51.445 [Info] (FastFlowLM) Loading model: gemma4-it-e2b-FLM
[FastFlowLM] Options: ctx_size=4096
2026-07-10 17:35:51.445 [Info] (FastFlowLM) Found flm at: flm
2026-07-10 17:35:51.463 [Error] (Router) Retry also failed: FLM NPU validation failed: NPU firmware is incompatible. Kernel version is incompatible.
2026-07-10 17:35:51.463 [Error] (Router) Visit https://lemonade-server.ai/driver_install.html for driver installation instructions.
2026-07-10 17:35:51.463 [Info] (FastFlowLM) Unloading model...
2026-07-10 17:35:51.463 [Info] (FastFlowLM) Unloading model...
2026-07-10 17:35:51.463 [Error] (Router) Failed to load model: FLM NPU validation failed: NPU firmware is incompatible. Kernel version is incompatible.
2026-07-10 17:35:51.463 [Error] (Router) Visit https://lemonade-server.ai/driver_install.html for driver installation instructions.
2026-07-10 17:35:51.463 [Error] (Server) Failed to load model: FLM NPU validation failed: NPU firmware is incompatible. Kernel version is incompatible.
2026-07-10 17:35:51.463 [Error] (Server) Visit https://lemonade-server.ai/driver_install.html for driver installation instructions.
2026-07-10 17:35:51.463 [Error] (Server) Error 500: POST /api/v1/load
系统 BUG:进入日志应用-更多-系统诊断日志,上传到社区

每次更新系统后,我可以执行我自己编译的一键脚本修复该BUG

修复文件我打包好了通过网盘分享的文件:AIMAX395飞牛NPU驱动修复
链接: https://pan.baidu.com/s/1BZRiPFuDuuE6Yl9BCRxffA?pwd=td6t 提取码: td6t

使用SSH登录主机后运行脚本。会运行NPU测试。得到如下结果

XRT
Version : 2.25.0
Branch : HEAD
Hash : 943586a79b5a714463cb13d3ba7e178b8532c817
Hash Date : Tue, 9 Jun 2026 19:31:23 -0700
amdxdna Version : 2.25.0_20260624, 86f1fafcacc0b58776ac96599a9eae790064272f
virtio-pci Version : 6.18.18-trim
NPU Firmware Version : 1.1.2.65

Device(s) Present

BDF Name Architecture Topology
[0000:c8:00.1] NPU Strix Halo aie2p 6x8

=== 恢复完成,无需重启!验证 NPU 性能 ===
source /opt/xilinx/xrt/setup.sh
xrt-smi validate --device 0000:c8:00.1

Autocomplete enabled for the xrt-smi command
XILINX_XRT : /opt/xilinx/xrt
PATH : /opt/xilinx/xrt/bin:/opt/xilinx/xrt/bin:/opt/xilinx/xrt/bin:/usr/local/bin:/usr/bin:/bin:/usr/local/games:/usr/games
LD_LIBRARY_PATH : /opt/xilinx/xrt/lib:/opt/xilinx/xrt/lib:/opt/xilinx/xrt/lib
PYTHONPATH : /opt/xilinx/xrt/python:/opt/xilinx/xrt/python:/opt/xilinx/xrt/python
WARNING: User doesn't have admin permissions to set performance mode. Running validate in default mode
Validate Device : [0000:c8:00.1]
Platform : NPU Strix Halo
Power Mode : default

Test 1 [0000:c8:00.1] : gemm
Details : TOPS: 51.0
Test Status : [PASSED]

Test 2 [0000:c8:00.1] : latency
Details : Average latency: 54.0 us
Test Status : [PASSED]

Test 3 [0000:c8:00.1] : throughput
Details : Average throughput: 73320.0 op/s
Test Status : [PASSED]

Validation completed

届时,这块AIMAX395的NPU可以单独运行小型模型,不占用GPU资源,温度功耗也很低,希望官方能配合显示NPU的UI那就更好了。

收藏
送赞
分享
您需要登录后才可以回帖 登录 | 立即注册

本版积分规则