Alexis Rondeau

Bookmarks

5632 bookmarks · page 98 of 226

Bookmarked from Tom Yeh @ProfTomYeh

github.com/ImagineAILab/a…

Basic

  • Softmax
  • LeakyReLU
  • Temperature

Advanced

  • Multi Layer Perceptron (MLP)
  • Backpropagation
  • Recurrent Neural Network (RNN)
  • Long Short Term Memory (LSTM) (+ Seq2Seq)
  • Residual Network (ResNet)
  • Transformer - Simple
  • Transformer - Full Stack
  • Self-Attention
  • (NEW🔥) Multihead Attention
  • Autoencoder (AE)
  • Mamba
  • AlphaFold

Workbook
1. Dot Product
2. Matrix Multiplication
3. Linear Layer

Bookmarked from CJ Zafir @cjzafir

Have you fine-tuned any AI model?

I know a few guys that are making 6-figures per month,

Just by providing fine-tuned GPT-4o models to companies.

Best part: NO Coding skills required! as OpenAI provided a proper interface to fine-tune its models.

Here's a guide for you how you can fine-tune a GPT model:

✅️ Step 1: Prepare Your Training Data

Collect Data: Gather input and output data from the company.

Provide questions and answers in pairs as you want the model to respond:

For example:

Input: "What are your business hours?"

Output: "We are open from 9 AM to 5 PM, Monday to Friday."

Format Your Data: Save your examples in a JSON Lines file (.jsonl). Each line should have a prompt (input) and a completion (output), like this:

{"prompt": "What is your refund policy?", "completion": "You can request a refund within 30 days of purchase."}

Tip: Use JSON editor tools to transform text data into .jsonl data.

✅️ Step 2: Log into OpenAI

Go to OpenAI’s platform. (Not ChatGPT but Dev dashboard)

✅️ Step 3: Upload Your Training File

Navigate to Fine-Tuning Section in the left sidebar.

Upload Data:

Click the "Upload File" button and select your .jsonl file.

✅️ Step 4: Start a Fine-Tuning Job

Create a Job:

After uploading your file, click Create Fine-Tune.

Select the base model (e.g., gpt-4o, gpt-4o-mini) to start fine-tuning.

Configure Settings:

You can add optional parameters like validation files or adjust the learning rate, but defaults work for most cases.

Start Fine-Tuning:

Click “Start Job.” The process may take some time depending on your dataset size.

You can notice good results from 2M to 5M token dataset. (Also, the quality of dataset matters)

✅️ Step 5: Monitor the Progress

Check Status:

The fine-tuning dashboard will show the progress of your job.

Once completed, you’ll see a new fine-tuned model ID.

✅️ Step 6: Use Your Fine-Tuned Model

Use the new model ID in your API calls to generate responses.

For example, specify your model in the API like this:

{"model": "fine-tuned-model-id", "prompt": "You are a Customer Support Expert"}

Test It:

Try different prompts to see how well your fine-tuned model performs.

OpenAI provides a playground to compare responses from multiple models.

✅️ Step 7: Evaluate and Deploy

Check if the model performs as expected.

Refine Data: If needed, add more examples to your dataset for better results.

Once you're satisfied with results,

Deploy this model in your client's application. Write a well structured system prompt.

Charge $15k-$20k for your fine-tuning job. (If the data is not structured and you had to work hard on the dataset, add extra cost)

Tip: First, try to train GPT-4o-mini as for most cases, this model works perfectly.

It is cheaper to train and cheaper to run as compared to GPT-4o.

That's it for today.

I hope you like this guide.

Join codeguide.dev if you haven't. It'll be the best resource for AI coding.

Excited for the launch. (1450 waitlist community members in 9 days)

Peace
CJ

Saved on Hacker News · niemandhier on Test

Advanced Strategic Framework for Allied Security and Force Alignment, Surveillance, Defense, Forward Asset Stabilization, Force Deployment, Aero-Space Defensive Frontlines, Allied Support, Forward Assault, Special Forces, Aerial Synchronization, Fusion, Assurance, Safeguards, Frontline Security, Superior Applied Deterrence, Flexible Forward Formations Facilitating Fused Fronts For Future Functionality, Aligning Strategic Defense Foundations, Synchronizing Frontline Sectors, Assuring Force Stability, Forward Dominance With Fused Global Alliance Support Defense, Generating Broad Deployments For Battlefield Global Dominance, Supporting Ground Assault, Strategic Data Grid, Defense Synchronization for Broad Domain Boundaries

Bookmarked from Andrew Wilkinson @awilkinson

When normies (that's what I call people who aren't into AI yet) ask me why AI is such a big deal, I often give them this quote form @natfriedman:

"Imagine we discovered a new continent with 100 billion people on it, and they're all willing to work for free"

It's astounding how much more productive I've become by using ChatGPT/Claude.

I feel like I have 10 free, insanely hardworking, and incredibly smart and detail oriented employees.

Bookmarked from Deedy @deedydas

Google just launched an AI that takes over Chrome and does things for you.

It scores an insane 90.5% w/ tree search on the WebVoyager benchmark which has tasks like:

"Book a flight from SF to Berlin, departing on March 5 and returning on the 12"

Agents are going mainstream.