LLM + LSTM = Large Memory Models (LMMs)

Many data scientists don't know how to push ML models to production. Here's the recipe 👇

𝗞𝗲𝘆 𝗜𝗻𝗴𝗿𝗲𝗱𝗶𝗲𝗻𝘁𝘀

🔹 𝗧𝗿𝗮𝗶𝗻 / 𝗧𝗲𝘀𝘁 𝗗𝗮𝘁𝗮𝘀𝗲𝘁 - Ensure Test is representative of Online data
🔹 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 - Generate features in real-time
🔹 𝗠𝗼𝗱𝗲𝗹 𝗢𝗯𝗷𝗲𝗰𝘁 - Trained SkLearn or Tensorflow Model
🔹 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗖𝗼𝗱𝗲 𝗥𝗲𝗽𝗼 - Save model project code to Github
🔹 𝗔𝗣𝗜 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 - Use FastAPI or Flask to build a model API
🔹 𝗗𝗼𝗰𝗸𝗲𝗿 - Containerize the ML model API
🔹 𝗥𝗲𝗺𝗼𝘁𝗲 𝗦𝗲𝗿𝘃𝗲𝗿 - Choose a cloud service; e.g. AWS sagemaker
🔹 𝗨𝗻𝗶𝘁 𝗧𝗲𝘀𝘁𝘀 - Test inputs & outputs of functions and APIs
🔹 𝗠𝗼𝗱𝗲𝗹 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴 - Evidently AI, a simple, open-source for ML monitoring

𝗣𝗿𝗼𝗰𝗲𝗱𝘂𝗿𝗲

𝗦𝘁𝗲𝗽 𝟭 - 𝗗𝗮𝘁𝗮 𝗣𝗿𝗲𝗽𝗮𝗿𝗮𝘁𝗶𝗼𝗻 & 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴

Don't push a model with 90% accuracy on train set. Do it based on the test set - if and only if, the test set is representative of the online data. Use SkLearn pipeline to chain a series of model preprocessing functions like null handling.

𝗦𝘁𝗲𝗽 𝟮 - 𝗠𝗼𝗱𝗲𝗹 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁

Train your model with frameworks like Sklearn or Tensorflow. Push the model code including preprocessing, training and validation scripts to Github for reproducibility.

𝗦𝘁𝗲𝗽 𝟯 - 𝗔𝗣𝗜 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 & 𝗖𝗼𝗻𝘁𝗮𝗶𝗻𝗲𝗿𝗶𝘇𝗮𝘁𝗶𝗼𝗻

Your model needs a "/predict" endpoint, which receives a JSON object in the request input and generates a JSON object with the model score in the response output. You can use frameworks like FastAPI or Flask. Containzerize this API so that it's agnostic to server environment

𝗦𝘁𝗲𝗽 𝟰 - 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 & 𝗗𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁

Write tests to validate inputs & outputs of API functions to prevent errors. Push the code to remote services like AWS Sagemaker.

𝗦𝘁𝗲𝗽 𝟱 - 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴

Set up monitoring tools like Evidently AI, or use a built-in one within AWS Sagemaker. I use such tools to track performance metrics and data drifts on online data.

Application of 1 bit LLM model

1️⃣ In a remote village, a student can use a mobile device with a 1-bit LLM to get personalized tutoring without internet access.

2️⃣ In a low-resource clinic, healthcare workers use a mobile app with a 1-bit LLM to diagnose common diseases from symptoms or images offline.

3️⃣ Farmers use a 1-bit LLM app to diagnose crop diseases and receive personalized farming advice based on soil type and weather patterns

4️⃣ In a disaster-prone area, a 1-bit LLM-powered app helps first responders and citizens communicate critical information in multiple languages offline

AI Agents are about to change everything—and it’s happening now.

Here’s the cheat sheet:
1️⃣ Agentic RAG Routers: Think of them as traffic controllers for your workflows.
2️⃣ Query Planning RAG: Perfect for making tasks super efficient.
3️⃣ Adaptive RAG: Always learning, always improving.
4️⃣ Corrective RAG: Spotting and fixing errors before they derail you.
5️⃣ Self-Reflective RAG: Basically, AI journaling to improve itself.
6️⃣ Speculative RAG: Solving problems before you even know they exist.
7️⃣ Self Route RAG: Dynamic workflow magic.

Everyone knows about LLM aka Large Language model.

Now we will talk about SLM aka Small Language model

As their name implies, SLMs are smaller in scale and scope than large language models.

Some examples of SLM are
- Phi 3.5
- tiny Llama
- mobile Llama
- Gemma2

SLMs can be trained using two main techniques:

Knowledge distillation: A smaller model learns from a larger, already-trained model

Pruning: Extra bits that aren't needed are removed to make the model faster and leaner

Here are some characteristics of SLMs:

Smaller in size: SLMs have fewer parameters than LLMs, often in the tens to hundreds of millions, compared to billions in LLMs.

More efficient: SLMs are more computationally efficient and can run on less powerful hardware.

Faster training: SLMs can be trained and developed faster than LLMs.

Specialized: SLMs are trained on curated data sources and can be specialized in specific tasks.

Fine-tunable: SLMs can be fine-tuned to do exactly what is needed for a specific task.

Cost-effective: SLMs can be more cost-effective than LLMs, making them a good option for integrating intelligent features when resources are limited.

You can join our whatsapp channel

https://whatsapp.com/channel/0029VaI2tnVFMqrThp6yEp1N

???? ???????? ?? Interview Experience at PayPal.

I wanted to share my experience interviewing for the ???? ???????? ?? position at PayPal.

Here's a breakdown of the process:

?????? ?????????? (??):
The first step was an online assessment sent by the recruiter. Clearing this assessment led to two technical rounds being scheduled, separated by a gap of five days.

????????? ????? ?:
This round was with a Data Engineer III and focused on problem-solving and SQL.

?). ??? ?????????:
1. ?ℎ? ????????? ???? ???????.
2. ? ???????? ????? ??????? (I don't recall the exact details but was similar to those dealing with task prioritization).

?). ??? ?????????:
Focused on window functions, their usage, and optimization strategies.

????????? ????? ? (?????? ?????):
This was done with a Staff Data Engineer and had three main parts:

A). ??????? ??????????:
Shared details about my past projects. Also discussed best practices for software and data engineering, including how I implemented these in my projects.

B). ?????? ????????:
The scenario involved multiple data sources such as Hadoop, S3, and Oracle DB. I was tasked with designing a solution to migrate data to a final S3 bucket.
Explained my choices for services and tools, including error logging, scalability, and fault tolerance.

C). ????? ?????? ?????????:
Given two data frames, I had to perform some processing and store the final output in another data frame.

?????????? ????? (????? ?):
This was with the Senior Engineering Manager, who was also the hiring manager for this role.

?????? ?????????:
A). ???????? : A deep dive into my projects, focusing on why specific tools and services were chosen.
B). ???? ???? ???????? :
How I would handle pipeline issues, like overload situations or service downtimes.
Behavioral Questions: Highlighted my problem-solving, teamwork, and adaptability skills.

?? ????? (????? ?):
The final round was with HR. We discussed the offer details PayPal was providing, covered some standard behavioral questions related to company culture and expectations.

Credit- Shubham shukla

https://dataspoof4081.graphy.com/membership