activatedgeek/LeNet-5 Updated 2025-01-01 +Created 1970-01-01
View more
Good packaging! Tested on Ubuntu 22.10:
git clone https://github.com/activatedgeek/LeNet-5
cd LeNet-5
git checkout 95b55a838f9d90536fd3b303cede12cf8b5da47f
virtualenv -p python3 .venv
. .venv/bin/activate

# Their requirements.txt uses >= and some == are incompatible with our Ubuntu.
pip install
  Pillow==6.2.0 \
  numpy==1.24.2 \
  onnx==1.13.1 \
  torch==2.0.0 \
  torchvision==0.15.1 \
  visdom==0.2.4 \
;

time python run.py
This throws a billion exceptions because we didn't start the visdom server, but never mind that.
The scrip does a fixed 15 epochs.
Output on P51:
real    2m10.262s
user    11m9.771s
sys     0m26.368s
The run also produces a lenet.onnx ONNX file, which is pretty neat, and allows us for example to visualize it on Netron:
Figure 1.
Netron visualization of the activatedgeek/LeNet-5 ONNX output
. From this we can see the bifurcation on the computational graph as done in the code at:
output = self.c1(img)
x = self.c2_1(output)
output = self.c2_2(output)
output += x
output = self.c3(output)
This doesn't seem to conform to the original LeNet-5 however?
Read the full article
Lenovo ThinkPad P51 (2017) log Updated 2025-01-01 +Created 1970-01-01
View more
  • battery life:
    • 2023-04: on-browser streaming + light browsing on Ubuntu 22.10: about 2h45. Too low! Gotta try buying a new battery.
  • 2022-01-04 updated firmward after noticing that ubuntu 21.10 does not wake up from suspend seemed to happen every time when not connected to external power. dmidecode diff excerpt:
     BIOS Information
        Vendor: LENOVO
-       Version: N1UET40W (1.14 )
-       Release Date: 09/28/2017
+       Version: N1UET71W (1.45 )
+       Release Date: 07/18/2018
    used the "Ubuntu Software" GUI as mentioned at: support.lenovo.com/gb/en/solutions/ht510810-how-to-do-software-updates-linux. Kudos for making this accessible to newbs.
    After doing that, another update became available to: 0.1.56, clicked it and was much faster than the previous one, and didn't auto reboot. After manual reboot, dmidecode diffed again:
     BIOS Information
        Vendor: LENOVO
-       Version: N1UET71W (1.45 )
-       Release Date: 07/18/2018
+       Version: N1UET82W (1.56 )
+       Release Date: 08/12/2021
    plus a bunch of other lines.
  • 2021-06-05 upgraded to Ubuntu 21.04 with a clean install from an ISO. Selected
    • "Minimal installation"
    • "Erase disk and install Ubuntu". Notably, this erased the Microsoft Windows that came with the computer and was never used not even once
    • "Erase disk ans use ZFS"
    • Encrypt the new Ubuntu installation for security
    After this, the GUI felt fast, who would have thought that erasing a bunch of stuff would make the system faster!
    lsblk contains:
    zd0               230:0    0   500M  0 disk
└─keystore-rpool  253:0    0   484M  0 crypt /run/keystore/rpool
nvme0n1           259:0    0 476.9G  0 disk
├─nvme0n1p1       259:1    0   512M  0 part  /boot/efi
├─nvme0n1p2       259:2    0     2G  0 part
│ └─cryptoswap    253:1    0     2G  0 crypt
├─nvme0n1p3       259:3    0     2G  0 part
└─nvme0n1p4       259:4    0 472.4G  0 part
    and lsblk -f:
    zd0               crypto_LUKS 2
└─keystore-rpool  ext4        1.0   keystore-rpool
nvme0n1
├─nvme0n1p1       vfat        FAT32
├─nvme0n1p2       crypto_LUKS 2
│ └─cryptoswap
├─nvme0n1p3       zfs_member  5000  bpool
└─nvme0n1p4       zfs_member  5000  rpoo
    Then:
    grep '[rb]pool' /proc/mounts
    contains:
    rpool/ROOT/ubuntu_uvs1fq / zfs rw,relatime,xattr,posixacl 0 0
rpool/USERDATA/ciro_czngbg /home/ciro zfs rw,relatime,xattr,posixacl 0 0
rpool/USERDATA/root_czngbg /root zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/srv /srv zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/usr/local /usr/local zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/games /var/games zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/log /var/log zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/lib /var/lib zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/mail /var/mail zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/snap /var/snap zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/www /var/www zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/spool /var/spool zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/lib/AccountsService /var/lib/AccountsService zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/lib/NetworkManager /var/lib/NetworkManager zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/lib/apt /var/lib/apt zfs rw,relatime,xattr,posixacl 0 0
rpool/ROOT/ubuntu_uvs1fq/var/lib/dpkg /var/lib/dpkg zfs rw,relatime,xattr,posixacl 0 0
bpool/BOOT/ubuntu_uvs1fq /boot zfs rw,nodev,relatime,xattr,posixacl 0 0
    which gives an idea of how the above map to mountpoints.
    Had two GUI freezes since installation, a fixed images shows no matter what I do, possibly graphics only, but impossible to tell (next time I'll try SSH access). No Nvidia drivers installed yet.
2020-06-06: dropped some lemon juice on the bottom left of touchpad. Bottom left button not working anymore... I'm an idiot. There are many other alternatives, but very aggravating, I'll replace it for sure. Can't find the exact replacement part or any videos showing its replacement online easliy, dang. For the T430: www.youtube.com/watch?v=F3lzV9uXRjU Asked at: forums.lenovo.com/t5/ThinkPad-P-and-W-Series-Mobile-Workstations/P51-left-bottom-button-below-trackpad-mouse-left-click-stopped-working-possible-to-replace/m-p/5019903 Also I could not access it because you need to remove the HDD first: www.youtube.com/watch?v=5Klawxc7T_Y and I can't pull it out even with considerable force, unlike in the video... And OMG, those button caps are impossible to re-install once removed!!! Then when I put the whole thing back together, the upper buttons were not working anymore. FUUUUUUUUCK. When first opening I pulled on it without properly removing the cap and it came off, but it didn't look broken in any way and I put it back in. Keyboard works thank God, so right black connector is fine, left white one oppears to be the one for upper keys and trackpoint, both of which stopped working. The hardware manual confirms that they are both part of the same device, so basically a mouse :-) TODO can it be bought separately from te keyboard? Doesn't look like it, photo of keyboard part includes those buttons. The manual also confirms that the bottom buttons are one device with the trackpad "trackpad with buttons", thus forming the second entire mouse.
2019-04-17: popup asking about "ThinkPad P51 Management Engine Update" from from 182.29.3287 to 184.60.3561, said yes.
Ubuntu 17.10 setup after buying it:
Battery life shown by Ubuntu battery app after installation:
  • before NVIDIA driver setup: 8h
  • after: 6.5h
Read the full article
@cirosantilli/_file/prime-number-theorem Updated 2025-01-01 +Created 1970-01-01
View more
Consider this is a study in failed computational number theory.
The approximation converges really slowly, and we can't easy go far enough to see that the ration converges to 1 with only awk and primes:
sudo apt intsall bsdgames
cd prime-number-theorem
./main.py 100000000
Runs in 30 minutes tested on Ubuntu 22.10 and P51, producing:
Figure 1.
Linear vs approximation plot
. and are added to give a better sense of scale. is too close to 0 and not visible, and the approximation almost overlaps entirely with .
Figure 2.
. It is clear that the difference diverges, albeit very slowly.
Figure 3.
. We just don't have enough points to clearly see that it is converging to 1.0, the convergence truly is very slow. The logarithm integral approximation is much much better, but we can't calculate it in awk, sadface.
But looking at: en.wikipedia.org/wiki/File:Prime_number_theorem_ratio_convergence.svg we see that it takes way longer to get closer to 1, even at it is still not super close. Inspecting the code there we see:
(* Supplement with larger known PrimePi values that are too large for \
Mathematica to compute *)
LargePiPrime = {{10^13, 346065536839}, {10^14, 3204941750802}, {10^15,
     29844570422669}, {10^16, 279238341033925}, {10^17,
    2623557157654233}, {10^18, 24739954287740860}, {10^19,
    234057667276344607}, {10^20, 2220819602560918840}, {10^21,
    21127269486018731928}, {10^22, 201467286689315906290}, {10^23,
    1925320391606803968923}, {10^24, 18435599767349200867866}};
so OK, it is not something doable on a personal computer just like that.
Read the full article
Go UI Updated 2025-01-01 +Created 1970-01-01
View more
On Ubuntu 22.10:
sudo apt install quarry
quary
just works with GNU go.
Read the full article
Hide top bar on Ubuntu Updated 2025-01-01 +Created 1970-01-01
View more
This has annoyed Ciro Santilli for many years, it is just too wasteful of screen space on laptops!
Or likely more generally, on GNOME desktop, which is the default desktop environment as of Ubuntu 22.10.
Read the full article
LLVM IR hello world Updated 2025-01-01 +Created 1970-01-01
View more
Example: llvm/hello.ll adapted from: llvm.org/docs/LangRef.html#module-structure but without double newline.
To execute it as mentioned at github.com/dfellis/llvm-hello-world we can either use their crazy assembly interpreter, tested on Ubuntu 22.10:
sudo apt install llvm-runtime
lli hello.ll
This seems to use puts from the C standard library.
Or we can Lower it to assembly of the local machine:
sudo apt install llvm
llc hello.ll
which produces:
hello.s
and then we can assemble link and run with gcc:
gcc -o hello.out hello.s -no-pie
./hello.out
or with clang:
clang -o hello.out hello.s -no-pie
./hello.out
hello.s uses the GNU GAS format, which clang is highly compatible with, so both should work in general.
Read the full article
Matter.js Updated 2025-01-01 +Created 1970-01-01
View more
To run the demos locally, tested on Ubuntu 22.10:
git clone https://github.com/liabru/matter-js
cd matter-js
git checkout 0.19.0
npm install
npm run dev
and this opens up the demos on the browser.
Read the full article
MLperf v2.1 ResNet Updated 2025-01-01 +Created 1970-01-01
View more
Ubuntu 22.10 setup with tiny dummy manually generated ImageNet and run on ONNX:
sudo apt install pybind11-dev

git clone https://github.com/mlcommons/inference
cd inference
git checkout v2.1

virtualenv -p python3 .venv
. .venv/bin/activate
pip install numpy==1.24.2 pycocotools==2.0.6 onnxruntime==1.14.1 opencv-python==4.7.0.72 torch==1.13.1

cd loadgen
CFLAGS="-std=c++14" python setup.py develop
cd -

cd vision/classification_and_detection
python setup.py develop
wget -q https://zenodo.org/record/3157894/files/mobilenet_v1_1.0_224.onnx
export MODEL_DIR="$(pwd)"
export EXTRA_OPS='--time 10 --max-latency 0.2'

tools/make_fake_imagenet.sh
DATA_DIR="$(pwd)/fake_imagenet" ./run_local.sh onnxruntime mobilenet cpu --accuracy
Last line of output on P51, which appears to contain the benchmark results
TestScenario.SingleStream qps=58.85, mean=0.0138, time=0.136, acc=62.500%, queries=8, tiles=50.0:0.0129,80.0:0.0137,90.0:0.0155,95.0:0.0171,99.0:0.0184,99.9:0.0187
where presumably qps means queries per second, and is the main results we are interested in, the more the better.
Running:
tools/make_fake_imagenet.sh
produces a tiny ImageNet subset with 8 images under fake_imagenet/.
fake_imagenet/val_map.txt contains:
val/800px-Porsche_991_silver_IAA.jpg 817
val/512px-Cacatua_moluccensis_-Cincinnati_Zoo-8a.jpg 89
val/800px-Sardinian_Warbler.jpg 13
val/800px-7weeks_old.JPG 207
val/800px-20180630_Tesla_Model_S_70D_2015_midnight_blue_left_front.jpg 817
val/800px-Welsh_Springer_Spaniel.jpg 156
val/800px-Jammlich_crop.jpg 233
val/782px-Pumiforme.JPG 285
where the numbers are the category indices from ImageNet1k. At gist.github.com/yrevar/942d3a0ac09ec9e5eb3a see e.g.:
  • 817: 'sports car, sport car',
  • 89: 'sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita',
and so on, so they are coherent with the image names. By quickly looking at the script we see that it just downloads from Wikimedia and manually creates the file.
TODO prepare and test on the actual ImageNet validation set, README says:
Prepare the imagenet dataset to come.
Since that one is undocumented, let's try the COCO dataset instead, which uses COCO 2017 and is also a bit smaller. Note that his is not part of MLperf anymore since v2.1, only ImageNet and open images are used. But still:
wget https://zenodo.org/record/4735652/files/ssd_mobilenet_v1_coco_2018_01_28.onnx
DATA_DIR_BASE=/mnt/data/coco
export DATA_DIR="${DATADIR_BASE}/val2017-300"
mkdir -p "$DATA_DIR_BASE"
cd "$DATA_DIR_BASE"
wget http://images.cocodataset.org/zips/val2017.zip
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
unzip val2017.zip
unzip annotations_trainval2017.zip
mv annotations val2017
cd -
cd "$(git-toplevel)"
python tools/upscale_coco/upscale_coco.py --inputs "$DATA_DIR_BASE" --outputs "$DATA_DIR" --size 300 300 --format png
cd -
Now:
./run_local.sh onnxruntime mobilenet cpu --accuracy
fails immediately with:
No such file or directory: '/path/to/coco/val2017-300/val_map.txt
The more plausible looking:
./run_local.sh onnxruntime mobilenet cpu --accuracy --dataset coco-300
first takes a while to preprocess something most likely, which it does only one, and then fails:
Traceback (most recent call last):
  File "/home/ciro/git/inference/vision/classification_and_detection/python/main.py", line 596, in <module>
    main()
  File "/home/ciro/git/inference/vision/classification_and_detection/python/main.py", line 468, in main
    ds = wanted_dataset(data_path=args.dataset_path,
  File "/home/ciro/git/inference/vision/classification_and_detection/python/coco.py", line 115, in __init__
    self.label_list = np.array(self.label_list)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 2 dimensions. The detected shape was (5000, 2) + inhomogeneous part.
TODO!
Read the full article
Stockfish CLI Updated 2025-01-01 +Created 1970-01-01
View more
Most of what follows is part of the Universal Chess Interface. Tested on Ubuntu 22.10, Stockfish 14.1.
After starting stockfish on the command line, d (presumably display) contains:
 +---+---+---+---+---+---+---+---+
 | r | n | b | q | k | b | n | r | 8
 +---+---+---+---+---+---+---+---+
 | p | p | p | p | p | p | p | p | 7
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 6
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 5
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 4
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 3
 +---+---+---+---+---+---+---+---+
 | P | P | P | P | P | P | P | P | 2
 +---+---+---+---+---+---+---+---+
 | R | N | B | Q | K | B | N | R | 1
 +---+---+---+---+---+---+---+---+
   a   b   c   d   e   f   g   h

Fen: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
Key: 8F8F01D4562F59FB
Sweet ASCII art. where:
Move white king's pawn from e2 to e4:
position startpos moves e2e4
Then display again:
d
gives:
 +---+---+---+---+---+---+---+---+
 | r | n | b | q | k | b | n | r | 8
 +---+---+---+---+---+---+---+---+
 | p | p | p | p | p | p | p | p | 7
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 6
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 5
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | P |   |   |   | 4
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 3
 +---+---+---+---+---+---+---+---+
 | P | P | P | P |   | P | P | P | 2
 +---+---+---+---+---+---+---+---+
 | R | N | B | Q | K | B | N | R | 1
 +---+---+---+---+---+---+---+---+
   a   b   c   d   e   f   g   h

Fen: rnbqkbnr/pppppppp/8/8/4P3/8/PPPP1PPP/RNBQKBNR b KQkq - 0 1
Key: B46022469E3DD31B
so we see that the pawn moved.
Now let's make Stockfish think for one second what is the next best move for black:
go movetime 1000
gives as the last line:
bestmove c7c5 ponder g1f3
TODO:
  • what is ponder? Something to do with thinking on the opponent's turn: permanent brain.
  • understand the previous lines
To make the move it as suggested for black, we have to either repeat the entire sequence of movements:
position startpos moves e2e4 c7c5
d:
 +---+---+---+---+---+---+---+---+
 | r | n | b | q | k | b | n | r | 8
 +---+---+---+---+---+---+---+---+
 | p | p |   | p | p | p | p | p | 7
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 6
 +---+---+---+---+---+---+---+---+
 |   |   | p |   |   |   |   |   | 5
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | P |   |   |   | 4
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   | 3
 +---+---+---+---+---+---+---+---+
 | P | P | P | P |   | P | P | P | 2
 +---+---+---+---+---+---+---+---+
 | R | N | B | Q | K | B | N | R | 1
 +---+---+---+---+---+---+---+---+
   a   b   c   d   e   f   g   h

Fen: rnbqkbnr/pp1ppppp/8/2p5/4P3/8/PPPP1PPP/RNBQKBNR w KQkq - 0 2
Key: 4CA78BCE9C2980B0
or alternatively we could also use the previous FEN notation as a starting point;
position fen rnbqkbnr/pppppppp/8/8/4P3/8/PPPP1PPP/RNBQKBNR b KQkq - 0 1 moves c7c5
Note how the Universal Chess Interface interface is very simple: we just load a state and then decide what to do next for that one state. The engine holds only one and exactly one state at a time, and you can't even modify it differentially without loading new one from scratch.
Let's move white again with our brain with either:
position startpos moves e2e4 c7c5 d2d3
position fen rnbqkbnr/pp1ppppp/8/2p5/4P3/8/PPPP1PPP/RNBQKBNR w KQkq - 0 2 moves d2d3
Set a specific position from fen:
position fen rnbqkbnr/pppppppp/8/8/4P3/8/PPPP1PPP/RNBQKBNR b KQkq - 0 1
Read the full article
torchvision ResNet Updated 2025-01-01 +Created 1970-01-01
View more
pytorch.org/vision/0.13/models.html has a minimal runnable example adapted to python/pytorch/resnet_demo.py.
That example uses a ResNet pre-trained on the COCO dataset to do some inference, tested on Ubuntu 22.10:
cd python/pytorch
wget -O resnet_demo_in.jpg https://upload.wikimedia.org/wikipedia/commons/thumb/6/60/Rooster_portrait2.jpg/400px-Rooster_portrait2.jpg
./resnet_demo.py resnet_demo_in.jpg resnet_demo_out.jpg
This first downloads the model, which is currently 167 MB.
We know it is COCO because of the docs: pytorch.org/vision/0.13/models/generated/torchvision.models.detection.fasterrcnn_resnet50_fpn_v2.html which explains that 
FasterRCNN_ResNet50_FPN_V2_Weights.DEFAULT
is an alias for:
FasterRCNN_ResNet50_FPN_V2_Weights.COCO_V1
The runtime is relatively slow on P51, about 4.7s.
After it finishes, the program prints the recognized classes:
['bird', 'banana']
so we get the expected bird, but also the more intriguing banana.
By looking at the output image with bounding boxes, we understand where the banana came from!
Figure 1.
python/pytorch/resnet_demo_in.jpg
. Source.
Figure 2.
python/pytorch/resnet_demo_out.jpg
. The beak was of course a banana, not a beak!
Read the full article
Twin prime conjecture Updated 2025-01-01 +Created 1970-01-01
View more
Let's show them how it's done with primes + awk. Edit. They have a -d option which also shows gaps!!! Too strong:
sudo apt install bsdgames
primes -d 1 100 | awk '/\(2\)/{print $1 - 2, $1 }'
gives us the list of all twin primes up to 100:
0 2
3 5
5 7
11 13
17 19
29 31
41 43
59 61
71 73
Tested on Ubuntu 22.10.
Read the full article
yolov5-pip Updated 2025-01-01 +Created 1970-01-01
View more
OK, now we're talking, two liner and you get a window showing bounding box object detection from your webcam feed!
python -m pip install -U yolov5==7.0.9
yolov5 detect --source 0
The accuracy is crap for anything but people. But still. Well done. Tested on Ubuntu 22.10, P51.
Video 1.
fcakyon/yolov5-pip webcam object detection demo by Ciro Santilli (2023)
Source.
Read the full article