New Tool: Maia Analysis

@Ironusmaidenus said ^

Nice tool! I am a big fan of the maia project. I wish I had a powerful GPU so I could integrate it into my local setup. Unfortunately, Lc0 is so slow on CPU it doesn't make sense for me to use it seriously.

The Maia1/2 models are very lightweight on CPU for just positional inference (i.e., not training or search). Maia3 is a little more expensive just due to parameter count, but still nearly instant inference on my 2022 Macbook Air.

What is your desired use case? Could you try this out and see if it is fast enough?

https://github.com/mcognetta/simple-maia3-inference

@fruitdealer2002 said ^

Nice tool! I am a big fan of the maia project. I wish I had a powerful GPU so I could integrate it into my local setup. Unfortunately, Lc0 is so slow on CPU it doesn't make sense for me to use it seriously.

The Maia1/2 models are very lightweight on CPU for just positional inference (i.e., not training or search). Maia3 is a little more expensive just due to parameter count, but still nearly instant inference on my 2022 Macbook Air.

What is your desired use case? Could you try this out and see if it is fast enough?

https://github.com/mcognetta/simple-maia3-inference

I am not sure when I will have some time to do this, I definitely will give it a shot at some point then give you an update. My initial plan was to add this to SCID vs PC so I can use it on positions where the engine is over my head. I am a beginner so I am trying to get away from masters databases and engine lines. I felt like Maia could provide a better alternative for my skill level. My final straw was when Stockfish told me I should sac my rook to remove a pawn defender! No way I could convert that to a win at my level.

@Ironusmaidenus said ^

Nice tool! I am a big fan of the maia project. I wish I had a powerful GPU so I could integrate it into my local setup. Unfortunately, Lc0 is so slow on CPU it doesn't make sense for me to use it seriously.

The Maia1/2 models are very lightweight on CPU for just positional inference (i.e., not training or search). Maia3 is a little more expensive just due to parameter count, but still nearly instant inference on my 2022 Macbook Air.

What is your desired use case? Could you try this out and see if it is fast enough?

https://github.com/mcognetta/simple-maia3-inference

I am not sure when I will have some time to do this, I definitely will give it a shot at some point then give you an update. My initial plan was to add this to SCID vs PC so I can use it on positions where the engine is over my head. I am a beginner so I am trying to get away from masters databases and engine lines. I felt like Maia could provide a better alternative for my skill level. My final straw was when Stockfish told me I should sac my rook to remove a pawn defender! No way I could convert that to a win at my level.

Oh nice, it's a simple Python script. This should be great for my use case even without direct integration into SCID. Thank you so much. I guess I should have researched more after my initial experience with Lc0 and Nibbler.

@Ironusmaidenus said ^

Nice tool! I am a big fan of the maia project. I wish I had a powerful GPU so I could integrate it into my local setup. Unfortunately, Lc0 is so slow on CPU it doesn't make sense for me to use it seriously.

The Maia1/2 models are very lightweight on CPU for just positional inference (i.e., not training or search). Maia3 is a little more expensive just due to parameter count, but still nearly instant inference on my 2022 Macbook Air.

What is your desired use case? Could you try this out and see if it is fast enough?

https://github.com/mcognetta/simple-maia3-inference

I am not sure when I will have some time to do this, I definitely will give it a shot at some point then give you an update. My initial plan was to add this to SCID vs PC so I can use it on positions where the engine is over my head. I am a beginner so I am trying to get away from masters databases and engine lines. I felt like Maia could provide a better alternative for my skill level. My final straw was when Stockfish told me I should sac my rook to remove a pawn defender! No way I could convert that to a win at my level.

Oh nice, it's a simple Python script. This should be great for my use case even without direct integration into SCID. Thank you so much. I guess I should have researched more after my initial experience with Lc0 and Nibbler.

Awesome. Good luck! Feel free to ping me if there are features you would like.

Also, you might like my other puzzle tool (also built on Maia):

https://lichess.org/@/fruitdealer2002/blog/blunderclinic-dont-blunder/oNSyIUBb

Worth checking out as well:

https://www.maiachess.com/analysis/

Note: there are more strength levels there, up to 2600

Nice tool and I really like the UI. Nice for the eyes.

Where can I find your source code? Maia is GPL licensed, so you also have to publish your source code.

Chao Em. I am full of admiration for all the amazing stuff you keep putting out! But this one time...

1. e4 and Nf6 is by far the most probable answer on all rating levels. Are you sure about that?

This sounds like the next chess revolution. LET'S GOOO! You're a legend for creating this!

Currently I have 2 questions:

1. I went here (https://chessboardmagic.com/maiaanalysis) and imported a game, but whenever I click on a move, the board does not show the position.

2. When using maiachess.com, why is it that when you make Maia stronger (for example, using Maia 600, and then using Maia 2600), that the "Moves by Rating" graph changes?