hello I hijack this a little bit …
once I type “lspci” it gave me following output on a pi 5
0001:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries BCM2712 PCIe Bridge (rev 30)
0001:01:00.0 Processing accelerators: Axelera AI Metis AIPU (rev 02)
0002:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries BCM2712 PCIe Bridge (rev 30)
0002:01:00.0 Ethernet controller: Raspberry Pi Ltd RP1 PCIe 2.0 South Bridge
how to activate the device and getting to work with my python3 script
Just enough to go to voyager sdk folder, type from this folder:
source venv/bin/activate
And run inference.py script with desired model and input
eeehhh it breaks at requirement pycairo==1.24.0 on python3.11.2
i`ll try this to met /dev/metis-1:1:0 device
Create Install Directory
In your home folder:
mkdir v1.2.5
cd v1.2.5/
Install PCIe driver
1) Download driver
wget https://software.axelera.ai/artifactory/axelera-apt-source/metis-dkms/metis-dkms_0.07.16_all.deb
2) Make sure dkms is installed
sudo apt-get install dkms
3) Then install the driver
sudo dpkg -i metis-dkms_0.07.16_all.deb
sudo modprobe metis
i`ll try this to met /dev/metis-1:1:0 device
Create Install Directory
In your home folder:
mkdir v1.2.5
cd v1.2.5/
Install PCIe driver
1) Download driver
wget https://software.axelera.ai/artifactory/axelera-apt-source/metis-dkms/metis-dkms_0.07.16_all.deb
2) Make sure dkms is installed
sudo apt-get install dkms
3) Then install the driver
sudo dpkg -i metis-dkms_0.07.16_all.deb
sudo modprobe metis
Yes, this will install drivers, but then if your OS is not Ubuntu 22, you need install docker and run voyager setup from it and install
my OS is raspberryOS eg. bookworm Debian
but the docker container hung a bit
<19/67] libz.so.1 -> libz.so.1.2.11
<19/67] libxxhash.so.0 -> libxxhash.so.0.8.1
<19/67] libresolv.so.2 -> libresolv.so.2
<19/67] libc_malloc_debug.so.0 -> libc_malloc_debug.so.0
<19/67] /usr/local/lib: (from /etc/ld.so.conf.d/libc.conf:2)
<19/67] /lib: (from <builtin>:0)
my OS is raspberryOS eg. bookworm Debian
but the docker container hung a bit
<19/67] libz.so.1 -> libz.so.1.2.11
<19/67] libxxhash.so.0 -> libxxhash.so.0.8.1
<19/67] libresolv.so.2 -> libresolv.so.2
<19/67] libc_malloc_debug.so.0 -> libc_malloc_debug.so.0
<19/67] /usr/local/lib: (from /etc/ld.so.conf.d/libc.conf:2)
<19/67] /lib: (from <builtin>:0)
I believe you are using SD card, for me actual setup was about 3 hours to complete, thats why I was switch to usb to SSD setup. Keep watching...
yes a SD card with 100MB/s read/write speed
so I must break and start it on another day …
got a piroman 5 case … must setup an ssd within the 5GBps USB port
but for the first try it is good
thank you
Yes i Write this to /boot/firmware/config.txt
dtparam=pciex1_gen=3
Hello all,
Our recommendation is that you start from the beginning, imaging your micro SD card using Raspberry Pi imager.
We have tested Raspberry Pi OS and Ubuntu 24.04.
Then use the step by step guide in: https://support.axelera.ai/hc/en-us/articles/26362016484114-Bring-up-Voyager-SDK-in-Raspberry-Pi-5
Please log all your steps and output so we can help debugging in case you have issues.
Note that our installation process was much shorter than 3 hours.
These are examples of the microSD cards we tested:
Maybe you can try to use any of these cards.
Let us know.
Kind regards,
Victor
I have ordered a m.2 nvme crucial p310
and will do a new setup this Weekend
will Test the third Option With:
dd if=/dev/(sdcard) of=/dev/sda bs=4M status=progress ; sync;
and modify the size of the Image with gparted to fit Full size of the ssd
well and booting from USB
howto let the configured docker run after a reboot?
to save work or else!
With docker you can run: one device …. But what is when you got two m.2 addon cards within one raspberry?
Preperate a raspberry with two m.2 slots
docker run -it --privileged -v /tmp:/tmp -v /lib/modules:/lib/modules -v /run:/run --device=/dev/dri --device=/dev/vchiq --group-add video -e DISPLAY=$DISPLAY --device /dev/metis0 --device /dev/metis1 --network=host --name=voyager-sdk-1.2.5 ubuntu:22.04
metis device not longer visible
could it be broken?
```
0002:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries BCM2712 PCIe Bridge (rev 30)
0002:01:00.0 Ethernet controller: Raspberry Pi Ltd RP1 PCIe 2.0 South Bridge
```
lspci -vvv
```
0002:00:00.0 PCI bridge: Broadcom Inc. and subsidiaries BCM2712 PCIe Bridge (rev 30) (prog-if 00 [Normal decode])
Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 0
Interrupt: pin A routed to IRQ 39
Bus: primary=00, secondary=01, subordinate=01, sec-latency=0
Memory behind bridge: 00000000-005fffff [size=6M] [32-bit]
Prefetchable memory behind bridge: 00000000fff00000-00000000000fffff [disabled] [64-bit]
Secondary status: 66MHz- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- <SERR- <PERR-
BridgeCtl: Parity- SERR- NoISA- VGA- VGA16- MAbort- >Reset- FastB2B-
PriDiscTmr- SecDiscTmr- DiscTmrStat- DiscTmrSERREn-
Capabilities: <access denied>
Kernel driver in use: pcieport
0002:01:00.0 Ethernet controller: Raspberry Pi Ltd RP1 PCIe 2.0 South Bridge
Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx+
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 0
Interrupt: pin A routed to IRQ 39
Region 0: Memory at 1f00410000 (32-bit, non-prefetchable) [size=16K]
Region 1: Memory at 1f00000000 (32-bit, non-prefetchable) [virtual] [size=4M]
Region 2: Memory at 1f00400000 (32-bit, non-prefetchable) [size=64K]
Capabilities: <access denied>
Kernel driver in use: rp1
```
had a nice turn with deepseek-r1
let the raspberry pi blown high … for cpu usage … it runs fast
could anyone tell me is that normal … that a raspberry can run
quick as possible?
DeepSeek-R1 | 7B | 4.7GB | ollama run deepseek-r1 |
DeepSeek, nice! Great work @florit!
I’d guess CPU overhead might spike during preprocessing, yeah. Probably fairly normal that?
So were you running DeepSeek on the Metis M.2?
So were you running DeepSeek on the Metis M.2?
hi @Spanner ...
so I hope and the Metis M.2 is not broken … because I did not see the m.2 card anymore on my raspberry pi
but I think when “llama3.3” Ready downloaded I can made one more test
it´s a 70B with around 42GB model
What PSU are you using @florit? I’d think the official Pi5 PSU would provide enough power, but it could be worth us investigating that too, if the M.2 is no longer visible.
Just to double check, you mentioned you’ve tried the Geekworm X1004 HAT (which doesn’t look like it supports PCIe Gen3, which is an issue with Metis); what’s the other HAT you’ve been testing with?
That’s a really nice piece of kit, actually!
It does suggest that it’s PCIe Gen3 compatible, and I bet the power supply is more than adequate. Did you have it working in this case initially?
Worked until i make Updates