Mr-lwd
Mr-lwd
I successfully run exo on jetson orin nx 16gb using "source install.sh", and "pip install -e ." in conda-python3.12 will occur the same error about "anaconda3/envs/exo/lib/libgcc_s.so: invalid ELF header"。 What's...
> 我在 jetson orin nx 16gb 上使用“source install.sh”成功运行了 exo，在 conda-python3.12 中“pip install -e ”将出现相同的错误“anaconda3/envs/exo/lib/libgcc_s.so： invalid ELF header”。 > > 此外，另一个错误是 “pynvml.NVMLError_NotSupported：不支持“，这是因为 Jetson 设备不支持”pynvml”。需要更改获取 jetson 设备的 gpu 和内存信息的函数，而且没有那么难。 try: with open("/proc/device-tree/compatible")...
Here comes a problem, After I downloaded the llama3.2-8B，It can be loaded in the memory. But it was killed then.
> 这里来了一个问题，我下载了 llama3.2-8B 之后，就可以加载到内存中了。但它在那时被杀死了。 I run exo on two nodes successfully, but the model seems to be loaded in memory twice or more. The inference time is too long, and...
Sorry to bother you, but I'd love to know the solution.
 I can successfully run exo on my devices (Jetson Orin Nano 8GB + Jetson Orin Nx 16GB)，however the inference speed is too slow (1~2 tokens per second for llama3.2:1b-8bit)....
I maybe figure out the reason why the speed is too slow. Jetson devices just support python3.10 and torch2.5~2.6, though they can create venv in python3.12, but the cuda is...
> HI, I had install python 3.12 in my Jetson orin nx 16G, and install exo OK，but could get 0(zero) TFLOPS; another is mac mini m4 32G may had 8.9...
As far as I know ,all the devices need to download the paramters of the same model, and devices run exo in different IP address. Each IP address can be...
> 请问你遇到了这个问题了吗 I meet the same problem