Cloud Yoda 😶‍🌫️

I #FOLLOWBACK. Googler. Full Stack Dad. AI practitioner. GenAI. Distributed Systems, SRE, Web 3.0, Scalability and Web. Wanderlust. All views mine.

[Cloud Yoda 😶‍🌫️ @ykaas]

@philhchen Thanks great advice, I would also recommend https://t.co/2PAbavHjJ6 to anyone wanting to kick start their career in applied ai.

[Cloud Yoda 😶‍🌫️ @ykaas]

https://t.co/Jqb6I8BHs1

[Cloud Yoda 😶‍🌫️ @ykaas]

Really good blog by Alisa on her job search. Here are the resources she used to study ML/LLM stuffs:

1. Stanford's "Language Modelling from Scratch" course: https://t.co/5yLdS1B4DM
(To understand the breadth of the field and keep a coherent picture in the mind)

2. After getting the breadth, she deep dived into concepts ONE at a time using blogs, papers, implementing things from scratch.

3. Implementing / debugging a transformer comes up so often in interviews. Turn it into muscle memory: https://t.co/mykcyp1TFs…

4. Do Leetcode https://t.co/1S1wzopSBL…

5. Other Learning resources she shared:
a. Self-Attention & Transformers: https://t.co/IoRJ8uEBR8…

b. The Illustrated GPT-2: https://t.co/WWYATavGwo…

c. Backpropagation
https://t.co/aJgPpySZr8

d. Introduction to Policy Gradient for LMs
https://t.co/zpTNYiSvQ6…

e. Lightweight Guide to understanding GRPO and RL principles
https://t.co/dByY3V0zOK…

f. How to Scale Your Model
https://t.co/PxAMOwLEVj

[Cloud Yoda 😶‍🌫️ @ykaas]

@alisawuffles Thanks for sharing, very inspiring. IMO https://t.co/fLvI78Fy1M is one of best resources out there to prepare for ai ml interviews, plus Stanford ai ml videos on YouTube and top 50 leet code.

[Cloud Yoda 😶‍🌫️ @ykaas]

3/ CyberPower Line Interactive/Offline UPS UT2200E - 2200VA / 1320W, 230V (May 2026 Purchase/ Brand new)
Lenovo ThinkVision S22-4e | 21.5" (60.45cm) FHD IPS 100Hz Ultraslim Monitor
Basic Keybord and Mouse

Local pickup from Bangalore only, no shipping.

[Cloud Yoda 😶‍🌫️ @ykaas]

Selling my custom built AI PC
MotherBoard: ASUS ProArt Z790-CREATOR WIFI
Processor: Intel(R) Core(TM) i9-14900K (3.20 GHz)
Memory: 128 GB Corsair Vengeance 6000Mhz DDR5 CL30 B (32GB x 4)
GPU: Zotac GeForce RTX 5090 Amp Infinity
SSD: 8TB Samsung 990 Pro Gen4 M.2 NVMe SSD (4TB x 2)

[Cloud Yoda 😶‍🌫️ @ykaas]

2/ CPU Cooler/Fan: Corsair AIO CF360 360mm
Power Supply: Corsair RM1200X-Shift 80+ Gold Modular
Cabinet: Corsair 4000D Frame Black

Additional parts in PC: Arctic MX-4 High Performance Thermal Paste (FOC) / Arctic P12 Max Non RGB - 3 Units

[Cloud Yoda 😶‍🌫️ @ykaas]

@kellabyte For large-scale clusters (up to 65,000 nodes), GKE replaced etcd with Google Cloud Spanner as the backend state store, while still exposing the etcd API for compatibility.
Spanner gives horizontal scalability, global distribution, and low-latency consistency without etcd’s limits

[Cloud Yoda 😶‍🌫️ @ykaas]

If you’re a builder who reads a lot of papers, you need to check out https://t.co/P7WKOpcyqi. It breaks down real-world applications. Making research sharing so much easier. ExplainThisPaper has quietly become part of my research workflow. ✨ #AI #ArXiv #Research #Paper #GenAI

[Cloud Yoda 😶‍🌫️ @ykaas]

The 'Hardware Winter' is here. 🥶 OpenAI, 40% supply buyouts, and why RAM prices won't drop until 2028.

Crazy episode

https://t.co/5Mxns1bHU5

[Cloud Yoda 😶‍🌫️ @ykaas]

This means you can flexibly obtain TPU capacity for defined-duration workloads like ML training, fine-tuning, batch inference, and benchmarking workloads all while receiving the discounted pricing benefits of DWS.

Get started today at https://t.co/3IxjubCRWV

[Cloud Yoda 😶‍🌫️ @ykaas]

It's now easier than ever to cost-efficiently train ML models using Google Cloud Tensor Processing Units (TPUs), because TPUs are now available via Dynamic Workload Scheduler (DWS) Flex start.

[Cloud Yoda 😶‍🌫️ @ykaas]

4/ and autoscaling configurations so that you can get started with an optimal deployment -- saving you weeks of qualification. All of this information is available with a simple API call: https://t.co/M6qg8AKTvG

[Cloud Yoda 😶‍🌫️ @ykaas]

One of the most important steps when productionizing ML model serving is to optimize performance -- tweaking your model server configuration to maximize throughput and minimize cost while achieving your desired latency.

#Kubernetes #mlmodel #googlecloud #GKE #modeltraining

[Cloud Yoda 😶‍🌫️ @ykaas]

3/ his means it's now super easy to get standardized benchmarking results showing the performance of your inference endpoints. On GKE, we use this benchmarking framework to generate benchmarking results across many combinations of accelerators, model servers, open weight models

[Cloud Yoda 😶‍🌫️ @ykaas]

2/ For anyone using Google Kubernetes Engine, this is now easier than ever. Together with others in the Kubernetes community, we just released an open source GenAI inference performance benchmarking tool: https://t.co/MmYttbQEp0

[Cloud Yoda 😶‍🌫️ @ykaas]

5️⃣ 𝗦𝗼𝗹𝘃𝗶𝗻𝗴 𝗗𝗼𝗺𝗮𝗶𝗻-𝗦𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗣𝗿𝗼𝗯𝗹𝗲𝗺𝘀 𝗨𝘀𝗶𝗻𝗴 𝗟𝗟𝗠𝘀 (𝗺𝘂𝘀𝘁 𝗳𝗼𝗿 𝗔𝗜 𝗔𝗱𝗼𝗽𝘁𝗲𝗿𝘀)
🔗 https://t.co/eS2nKTleeP

6️⃣ 𝗠𝗟𝗢𝗽𝘀 𝗳𝗼𝗿 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜
🔗 https://t.co/7GcXNidkEc

[Cloud Yoda 😶‍🌫️ @ykaas]

𝗧𝗵𝗲𝘀𝗲 6 𝘄𝗵𝗶𝘁𝗲 𝗽𝗮𝗽𝗲𝗿𝘀 𝗳𝗿𝗼𝗺 Google 𝗮𝗿𝗲 𝗮 𝗴𝗼𝗹𝗱 𝗺𝗶𝗻𝗲

I often refer to them every now and then

If you're trying to Adopt AI in your business, I'd highly recommend you to go through these at least once

[Cloud Yoda 😶‍🌫️ @ykaas]

3️⃣ 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 & 𝗩𝗲𝗰𝘁𝗼𝗿 𝗦𝘁𝗼𝗿𝗲𝘀
🔗 https://t.co/7FAsHpokYf

4️⃣ 𝗔𝗴𝗲𝗻𝘁𝘀 (𝗠𝘆 𝗳𝗮𝘃𝗼𝘂𝗿𝗶𝘁𝗲 𝗼𝗻𝗲)
🔗 https://t.co/H9Kwf4UhXp

[Cloud Yoda 😶‍🌫️ @ykaas]

1️⃣ 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗟𝗟𝗠 𝗠𝗼𝗱𝗲𝗹𝘀 𝗮𝗻𝗱 𝗧𝗲𝘅𝘁 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻
🔗 https://t.co/rugpQOJo7s

2️⃣ 𝗣𝗿𝗼𝗺𝗽𝘁 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴
🔗 https://t.co/GFv951stEQ