Preet Rank

21, figuring out code, AI/ML & LLMs.
currently learning LLM inference, looking for an internship.

[Preet Rank @preettrank]

I’ve started a 12week journey into LLM inference.

The goal isn’t to collect courses or papers.

I want to understand what happens between a prompt and a generated token - then build, benchmark and optimize the system myself.

I’ll document the wins, failures and numbers here.

[Preet Rank @preettrank]

I recently dove into the Model Context Protocol.

First, I learned the basics by reading the docs.
Then, I set up existing servers and tested them.
Next, I built my own server using Python.
Finally, I debugged and deployed it.

Want to start your own?
https://t.co/HQodoeCvQN

[Preet Rank @preettrank]

I recently dove into the Model Context Protocol.

First, I learned the basics by reading the docs.
Then, I set up existing servers and tested them.
Next, I built my own server using Python.
Finally, I debugged and deployed it.

Want to start your own?
https://t.co/HQodoeCvQN

[Preet Rank @preettrank]

So apparently, if you learn LLM inference super fast, you’ll be ready for the future. Sounds easy, right?

[Preet Rank @preettrank]

So apparently, if you learn LLM inference super fast, you’ll be ready for the future. Sounds easy, right?

[Preet Rank @preettrank]

I’m looking for an off-campus internship in AI or AI engineering.

If anyone knows how to apply or has a guide, I’d really appreciate your help.

Please drop your tips or links.

Let’s support each other in this journey!

[Preet Rank @preettrank]

I’m looking for an off-campus internship in AI or AI engineering.

If anyone knows how to apply or has a guide, I’d really appreciate your help.

Please drop your tips or links.

Let’s support each other in this journey!

[Preet Rank @preettrank]

@abinav_gt Thanks man!

[Preet Rank @preettrank]

Before jumping into Transformers, I'd understand:

Linear regression
Logistic regression
Decision trees
Bias vs variance
Gradient descent
Regularization
Cross-validation
Feature scaling
Classification metrics
Probability

Deep learning makes much more sense afterward.

[Preet Rank @preettrank]

just because it talks doesn't mean it's smart

[Preet Rank @preettrank]

Learning LLM inference is key for AI engineers. It helps us build smarter models and solve real problems. Let’s dive in and explore this exciting area together. The future of AI is bright, and we can be part of it!

[voxium @v0xium]

The 20 papers I’d read/implement to become a better LLM inference engineer.

1. Attention Is All You Need
2. FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
3. FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning
4. FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision
5. Orca: A Distributed Serving System for Transformer-Based Generative Models
6. Efficient Memory Management for Large Language Model Serving with PagedAttention
7. AlpaServe: Statistical Multiplexing with Model Parallelism for Deep Learning Serving
8. SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models
9. GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
10. AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration
11. Fast Inference from Transformers via Speculative Decoding
12. SpecInfer: Tree-based Speculative Inference and Verification for Large Language Models
13. SGLang: Efficient Execution of Structured Language Model Programs
14. SARATHI-Serve: Taming Throughput–Latency Tradeoff in LLM Inference with Sarathi-Serve
15. Splitwise: Efficient Generative LLM Inference Using Phase Splitting
16. DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Serving
17. Mooncake: A KVCache-centric Disaggregated Architecture for LLM Serving
18. Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads
19. EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty
20. DeepSeek-V3 Technical Report

Some of these papers were recommended to me by an engineer working at NVIDIA, and I take his recommendations seriously. If possible, I would have loved to form a community where we could share implementations and learn from one another.

[Preet Rank @preettrank]

I am genuinely emotional looking at my dashboard today.

My post just hit over 500+ impressions first time 😇!

Small wins are still wins.

Let's go!

Target: 50 followers by the end of this month.

[Preet Rank @preettrank]

Before jumping into Transformers, I'd understand:

Linear regression
Logistic regression
Decision trees
Bias vs variance
Gradient descent
Regularization
Cross-validation
Feature scaling
Classification metrics
Probability

Deep learning makes much more sense afterward.

[Preet Rank @preettrank]

GN 😴

[Preet Rank @preettrank]

If LLMs feel overwhelming, you're probably learning them in the wrong order.

Master the Transformer first:

(Sort by impact, not hype)

Attention
Tokenization
Embeddings + position
FFN + residuals + normalization
Training + loss
KV cache + inference
Scaling + optimization

[Preet Rank @preettrank]

If LLMs feel overwhelming, you're probably learning them in the wrong order.

Master the Transformer first:

(Sort by impact, not hype)

Attention
Tokenization
Embeddings + position
FFN + residuals + normalization
Training + loss
KV cache + inference
Scaling + optimization

[Preet Rank @preettrank]

Before jumping into Transformers, I'd understand:

Linear regression
Logistic regression
Decision trees
Bias vs variance
Gradient descent
Regularization
Cross-validation
Feature scaling
Classification metrics
Probability

Deep learning makes much more sense afterward.

[Preet Rank @preettrank]

Before jumping into Transformers, I'd understand:

Linear regression
Logistic regression
Decision trees
Bias vs variance
Gradient descent
Regularization
Cross-validation
Feature scaling
Classification metrics
Probability

Deep learning makes much more sense afterward.

[Preet Rank @preettrank]

@Tushar26T Hehe

[Preet Rank @preettrank]

If you're learning AI/ML, don't skip these:

1. Train/validation split
2. Overfitting
3. Regularization
4. Feature engineering
5. Loss functions
6. Gradient descent
7. Learning rate
8. Batch size
9. Evaluation metrics
10. Data leakage

These matter more than the latest framework.

[SUPERBRO @Rushu_Tushu]

Is this the best channel to start learning AIML?

[Preet Rank @preettrank]

Day 2 of learning LLM inference.

Today I stopped studying and started building Micro-Transformer from Scratch.

I took 1.1M characters of Shakespeare and turned them into something a Transformer can actually learn from.

Here's what I built 👇

[Preet Rank @preettrank]

If you're learning AI/ML, don't skip these:

1. Train/validation split
2. Overfitting
3. Regularization
4. Feature engineering
5. Loss functions
6. Gradient descent
7. Learning rate
8. Batch size
9. Evaluation metrics
10. Data leakage

These matter more than the latest framework.

[Preet Rank @preettrank]

The LLM inference stack I'd study in 2026:

1. KV caching
2. Prefix caching
3. Continuous batching
4. Paged KV
5. Chunked prefill
6. Speculative decoding
7. Quantization
8. CUDA graphs
9. Async scheduling
10. Prefill/decode disaggregation