frutik

Machine Learning Scientist II - Content Intelligence

About the job
At Booking.com, data drives our decisions. Technology is at our core and innovation is everywhere. But our company is more than datasets, lines of code or A/B tests. We’re the thrill of the first night in a new place. The excitement of the next morning. The friends you make. The journeys you take. The sights you see. And the food you sample. Through our products, partners and people, we can empower everyone to experience the world.

The Content Intelligence team builds the Content Intelligence Platform by consuming millions of images and textual inputs every day, and then enriching it with ML capabilities. Eventually, these will serve downstream applications and personalize our customers' experience (think of a way to choose and surface the right images and reviews when customers book their next vacation).

Moreover the team is taking a key role in building in-house LLMs for different needs as: moderation, translation, AI trip planner chatbot

frutik

/**
 * Levenshtein distance between two strings.
 * - O(m*n) time
 * - O(min(m,n)) memory
 */
function levenshtein(a, b) {
  if (a === b) return 0;
  if (!a) return b.length;
  if (!b) return a.length;

frutik

Performance implications of medium size values and TOAST in Postgres and how to mitigate them - https://pganalyze.com/blog/5mins-postgres-TOAST-performance
The Surprising Impact of Medium-Size Texts on PostgreSQL Performance. Why TOAST is the best thing since sliced bread  - https://hakibenita.com/sql-medium-text-performance
Speed up your queries by avoiding to hit the TOAST - https://medium.com/@walttonm/speed-up-your-queries-by-avoiding-to-hit-the-toast-a71feaaeacc2

frutik

from mlx_lm import load, generate

model, tokenizer = load('Qwen/Qwen2-7B-Instruct-MLX', tokenizer_config={"eos_token": "<|im_end|>"})

prompt = "Why people call putin khuilo."
messages = [
    {"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(

frutik

CREATE OR REPLACE FUNCTION hamming_distance(
 A0 bigint, A1 bigint, A2 bigint, A3 bigint,
 B0 bigint, B1 bigint, B2 bigint, B3 bigint
)
RETURNS integer AS $$
BEGIN
RETURN 
    bits_count(A0 # B0) + 
    bits_count(A1 # B1) +
    bits_count(A2 # B2) + 

frutik

SELECT
    c.relname AS index_name,
    pg_size_pretty(pg_relation_size(c.oid)) AS index_size,
    pg_size_pretty(pg_total_relation_size(c.oid) - pg_relation_size(c.oid)) AS index_bloat_size,
    ROUND((pg_total_relation_size(c.oid) - pg_relation_size(c.oid)) / pg_relation_size(c.oid)::numeric * 100, 2) AS bloat_percentage
FROM
    pg_class c
JOIN
    pg_namespace n ON c.relnamespace = n.oid
WHERE

frutik

import onnxruntime as ort
from transformers import AutoTokenizer

session = ort.InferenceSession('./bge-small-en/model.onnx')
tokenizer = AutoTokenizer.from_pretrained("./bge-small-en")

inputs = tokenizer("hello world.", padding="longest", return_tensors="np")

inputs_onnx = {key: ort.OrtValue.ortvalue_from_numpy(value) for key, value in inputs.items()}

frutik

https://brandur.org/fragments/postgres-partitioning-2022
https://pganalyze.com/blog/postgresql-partitioning-django
https://django-postgres-extra.readthedocs.io/en/master/table_partitioning.html
https://hevodata.com/learn/postgresql-partitions/
https://www.2ndquadrant.com/en/blog/postgresql-12-foreign-keys-and-partitioned-tables/
https://www.postgresql.org/docs/current/ddl-partitioning.html#DDL-PARTITIONING-DECLARATIVE-LIMITATIONS

frutik

def divide_chunks(l, n):
    for i in range(0, len(l), n):
        yield l[i:i + n]

a = 'i.strip() for i in a.split('.') if i.strip()]
c = list(divide_chunks(b, 3))
d = ['. '.join(i + ['']).strip() for i in c]
y = '\n\n'.join(d)
print(y)

frutik

kubectl get --raw "/apis/metrics.k8s.io/v1beta1/nodes" -v=8 | python -m json.tool