Tried claude for the first time and I must say, I’m very impressed. The response from the model and as well as the UX of the application is very superior and easy to use. The artifact window is very useful. It is faster than I expected.

Now I’m going to try llama 3.2 1B model. Let’s see…It is very nice. I’ve also tried llama 3.2 3B model also, I’m very happy to run these models. These are faster than I expected. I used the model with zed and it tried to answer my questions (not very accurate).

I am planning to run a fine-tuning process with axolotl.

Found these,

• https://github.com/simonw/llm/tree/main
• https://datasette.io/tools
• https://github.com/simonw/files-to-prompt

In the past few days, I’ve exposed to a new term data locality. That is something do with CPU registers and stuff that improves code performance. This says that how we are coding matters in performance critical applications. But I’m not sure about day to day ones but it is good to know these stuff.

I came across this article now https://hackernoon.com/solving-the-all-pairs-shortest-paths-problem-with-the-floyd-warshall-algorithm-in-c and I love the way the author explains the approach and code. I couldn’t understand all the stuff (some of the concepts are very new to me) but I could grasp why it is done.

Found another good resource,

• https://christianjmills.com/posts/mastering-llms-course-notes/workshop-002/