@b0rk.jvns.ca Re unbuffering `tail | grep`, I used to use Tcl's Expect to do exactly that:
```
#!/usr/bin/expect
spawn bash -c ""
expect {
-re "PATTERN\n" {
puts "$expect_out(1,string)"
exp_continue
}
}
```
Not sure this still works.
```
#!/usr/bin/expect
spawn bash -c "
expect {
-re "PATTERN\n" {
puts "$expect_out(1,string)"
exp_continue
}
}
```
Not sure this still works.
December 3, 2024 at 7:18 PM
@b0rk.jvns.ca Re unbuffering `tail | grep`, I used to use Tcl's Expect to do exactly that:
```
#!/usr/bin/expect
spawn bash -c ""
expect {
-re "PATTERN\n" {
puts "$expect_out(1,string)"
exp_continue
}
}
```
Not sure this still works.
```
#!/usr/bin/expect
spawn bash -c "
expect {
-re "PATTERN\n" {
puts "$expect_out(1,string)"
exp_continue
}
}
```
Not sure this still works.
This is pretty cool: a @containerd.dev shim to run LLMs with ollama. The shim augments the image by mounting in the ollama binary, generating a Modelfile as input to ollama (based on the model bundled in the image), and setting an entry point.
![Mount Ollama binary
• Mount the Ollama binary from the host into the container
config. Mounts
append(config. Mounts, specs. Mount
Source:
"/usr/bin/ollama"
Destination:
"/usr/bin/ollama",
Type:
"bind".
Options:
[]string("rbind", "ro"},
})](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:e6d6ygltadw434kgrsfnjwql/bafkreifgp5livpbguqbhqp7otfuasyvlbaa377pxsggtoivtofdp3fj2eq@jpeg)
![Mount OCI as a Modelfile
• Ollama needs to read the GGUF from a Modelfile file
modelPath: getTaskEnv ("OLLAMA_MODEL PATH", config.Process.Env)
modelfile: fmt.Sprintf("FROM %s", modelPath)
err = os.WriteFile(filepath.Join(tempDir, "Modelfile"), []byte (modelfile), 0644)
if err != nil {
return nil, err
}
config.Mounts append(config.Mounts, specs. Mount(
Source:
filepath.Join(tempDir, "Modelfile"),
Destination: "/tmp/Modelfile",
Type:
"bind",
Options: []string("rbind", "ro"},
})](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:e6d6ygltadw434kgrsfnjwql/bafkreifml3eg2dvhqdknyruiowdyyxwpd4hoy4yiaw6bojsuaohx7vsjie@jpeg)
![Serve the Model
Set the container args to serve the model using Ollama
config.Process.Args = [] string(
/bin/bash",
"-c".
strings.Join([]string{
"ollama serve &",
"while ollama ps> /dev/null 2>&1; do sleep 1; done",
"ollama create modelName + --file/tmp/Modelfile",
"tail -f /dev/null",
}, "\n"),](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:e6d6ygltadw434kgrsfnjwql/bafkreigry7cnnqcywkk4tkhwpc3izqhjvuq2hm2xsvl7mjzjkxqitisuyu@jpeg)
April 2, 2025 at 3:40 PM
This is pretty cool: a @containerd.dev shim to run LLMs with ollama. The shim augments the image by mounting in the ollama binary, generating a Modelfile as input to ollama (based on the model bundled in the image), and setting an entry point.
text classifier implemented in bash
![A source code listing. The source code is written in bash and reads:
#!/usr/bin/env bash
deflate () {
echo "$1" | gzip -c
}
ncd () {
cx=$(deflate "$1" | wc -c)
cy=$(deflate "$2" | wc -c)
cxy=$(deflate "$1$2" | wc -c)
bc -l <<< "($cxy - $((cx < cy ? cx : cy)))/$((cx > cy ? cx : cy))"
}
knn () {
k=$1
test_=$2
trains=$3
dists=()
i=0
for train_label in "${trains[@]}"; do
((i++))
IFS='|||' read -r label train <<< "$train_label"
echo -en "classifying $i/120000...\r" >&2
dists+=("$(ncd $test_ $train)|||$label")
done
sorted=($(printf '%s\n' "${dists[@]}" | sort -n -t '|' -k1,1))
top_k=("${sorted[@]:0:$k}")
predicted_label=$(for e in "${top_k[@]}"; do
echo "$e" | awk -F'\|\|\|' '{print $2}'
done | sort | uniq -c | sort -nr | head -n1 | awk '{print $2}')
echo "$predicted_label"
}
parse () {
declare -n data=$1
while IFS=',' read -r label title description; do
data+=("$label|||$title $description")
done < <(tail -n +2 "$2" | head -n 100)
}
parse trains "train.csv"
echo -en "parsed training data\r"
parse tests "test.csv"
echo -en "parsed test data \r"
correct=0
wrong=0
for test_ in "${tests[@]}"; do
IFS='|||' read -r label text <<< "$test_"
predicted=$(knn 2 $text $trains)
if [ "$predicted" = "$label" ]; then
((correct++))
else
((wrong++))
fi
done
echo ""
echo "correct: $correct, wrong: $wrong"](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:3il2d6bhdt27woeaa6ba3f4n/bafkreifby2xiqqkl5xgzm777qq4po2cuna6nvbhrep3ek5gcmvghp27u6e@jpeg)
July 15, 2025 at 12:19 PM
text classifier implemented in bash
Hello world in 12 bytes of tail code:
$ cat a.tail
#!/usr/bin/tail -1
hello world
$ ./a.tail
hello world
$
But it’s wasteful because it uses an entire byte to represent just one letter.
$ cat a.tail
#!/usr/bin/tail -1
hello world
$ ./a.tail
hello world
$
But it’s wasteful because it uses an entire byte to represent just one letter.
Pfft I could write hello world in Python in 22 bytes. 🙄
February 20, 2025 at 7:12 PM
Hello world in 12 bytes of tail code:
$ cat a.tail
#!/usr/bin/tail -1
hello world
$ ./a.tail
hello world
$
But it’s wasteful because it uses an entire byte to represent just one letter.
$ cat a.tail
#!/usr/bin/tail -1
hello world
$ ./a.tail
hello world
$
But it’s wasteful because it uses an entire byte to represent just one letter.
