LLM Engineering course : Day 2 - 5

-

Open Source Vs Closed Source Model

 

 

See below , canvas is used for Coding along side with ChatGPT

 

 

 

 

 

Please note gap between words may come up as part of token , another example below

 


Rule of thumb is a token maps to 4 characters

Context Window 

 

LLM appear to have a memory , what is really happening is everytime you talk to LLM / ChatGPT the entire conversation (including initial ques , its response, followup ques etc) are passed in again as a long prompt. To understand what is most likely to come next.

What is Context Window ?

https://www.ibm.com/think/topics/context-window

 

Comparing different context windows

https://medium.com/@genai.works/comprehensive-comparison-of-large-language-models-llms-0da7e894e419

  

Compare Models

 

  •  gemini has context window of 1million token = 750k words = entire work of shakespear

What is one shot prompting ?

In the prompt we send the model, we give one example of what we looking for, the kind of thing we expecting it to reply.

Multi shot prompting = tell what to respond in different situations 

 

 

 

 

Comments

Post a Comment

Popular posts from this blog

Agentic AI Course : Week 1

-
Image
8 weeks https://edwarddonner.com/2024/11/13/llm-engineering-resources/  6 week Agentic AI Course https://github.com/ed-donner/agents the file .env has all the keys select kernel https://github.com/ed-donner/agents/blob/main/1_foundations/1_lab1.ipynb import os openai_api_key = os.getenv( 'OPENAI_API_KEY' ) from openai import OpenAI openai = OpenAI() messages = [{ "role" : "user" , "content" : "What is 2+2?" }] response = openai.chat.completions.create(      model= "gpt-4.1-nano" ,      messages=messages      ) print (response.choices[ 0 ].message.content) Day 2 : what are AI Agents? 5 different patterns for agentic systems https://lavnish.blogspot.com/2025/05/essential-design-patterns-for-building.html In General Day 3 comparing different models  https://www.vellum.ai/llm-leaderboard syntax to call diff LLMs openAI , anthropic, gemini, deepseek, groq https://github.com/ed-donner/agents/blob/main/1_foundations/2_lab2.ipynb...
Read more

LLM Engineering course : Day 1

-
Image
 Anaconda Power Shell    this will open , way you know if it is a anaconda power shell by seeing "base" Then you can install the full environment    then to activate use command "conda activate llms"     Open Ques "what is environment.yml" Then activate the conda environment and open jupyter lab   you can start with virtualenv as well (all details in readme) Video 11 : Day 1 every AI has 2 types of plan 1) pro plan with UI monthly subscription 2) API pricing , pay per request, for OPEN API need to put in 5$ pre payment (can use gpt4-o-mini for less cost)  (all details in readme)  Video 11:  store keys in .env files (all details in readme)  Video 14 there are two types of prompt   Video 15 OpenAI came up with a standard way of interacting with LLMs which is now standard for many LLMs. It is a list of dictionaries. Use completions API https://platform.openai.com/docs/guides/completions   For websites that use Java...
Read more

LLM Engineering : Week 2

-
Day 2.1 setup  anthropic api key google api key (more complex) update .env file     What is Temperature ?  Pass value between 0 to 1. 1 is most creative , 0 is least There is a slight difference in Claude API and OpenAI  in claude API we pass system message and user message separately in claude we also give context  if you want to stream , in claude there is an extra method , in openAI there is a boolean attribute.    Day 2.2 : Gradio         Day 2.3         Day 2.4 Day 2.5          
Read more