What happened to this project, is the library and freebasic .bi file still available.

Reading elsewhere some more recent posts about this project, bugs were encountered.

Interesting result when training, then testing, via python3 with torch. Indicates that once the Average Loss reaches a particular value, the NN will faithfully perform 4bit signed multiplication. We're also using thresholding after the NN output. The NN structure is: 8,128,16,8 and the activation fu...

Just ran ollama run qwen2.5-coder:1.5b this is fairly smooth, producing sensible freebasic code and glsl code that may also be sensible. At the moment these are just small pieces of code. If this is running completely upon my PC then this is remarkable. The only discomfort is over copyright, Alibaba...

There's also freebasic documentation as a text file, in zip format, that's available. I managed to install ollama, that runs a selection known as a manifest from various A.I groups, this has the ability to use VRAM and system RAM, according to the brief note I read. CUDA acceleration is possible, te...

Okay, so some time has passed and there have been some developments, like DeepSeek. However, at the practical level I'm interested in an online ai chat client where I don't need to worry about what I develop being visible to others, Meta's llama models don't meet that criterion. Also, I don't want t...

Part of the larger project, this reads Neural Network weights and biases from a text file. I've used a pytorch equivalent of the freebasic Neural Network code, with some interesting results. From there I saved the weights and biases as a text file, the same should also be possible from the previous ...

Using this structure for the NN, I progress through the loss like this. Upon testing the NN, the Total Error: 0.01267549334836566, is quite good; however, it needs to be 0. More training epochs and a further tweak to the NN structure might get me there. Dim layer_sizes(4) As Integer = {8, 64, 16, 16...

This is the present status of the NN for logic emulation. The use of randomised shuffling of indexes for the input and corresponding training data is recommended. The Sigmoid activation function may not be optimal. Temporary storage of weights and biases in auxilary arrays, then copy back depending ...

Previous, valid for 4bit and 8bit signed integers

As usual, once I get a flood of feedback I start coding. I might also examine this for 8 bit signed integers, anyway here's the code. ' ' ---------------------------------------------------------------------- ' int_bit3.bas ' ' signed integer to binary representation ' and ' binary representation to...

This is what I'm attempting, illustrated via code. ' NN_Logic.bas ' ---------------------------------------------------------------------- ' ' NN for Logic simulation . ' ' sciwiseg@gmail.com ' ' ---------------------------------------------------------------------- ' screen 12 color 15,0 locate 2,2...

A somewhat, related piece of code, requiring improvement. The type of routine one might make into a standard library, if it doesn't already exist. 'declare sub int2bit(bsq as Matrix, x as uinteger, nb as integer) declare Function BitsToInt(bits() As UShort , size As Integer) As Integer declare funct...

This is in preparation for something other than logic gates, which don't always reach levels 0 or 1. You can always use the previous code when you're assuming a definite 0 or 1, off or on. The introduction of MSE loss during training might be useful even in that context. I appreciate the value of ke...

Being a hot evening at my location, my brain isn't ready for too much coding. So, I asked DeepSeek to improve upon the existing code, this is the result, you'll note that the inference stage uses inputs ranging from 0 to 2. /' DS_NN2b.bas (c) Copyright 2025, sciwiseg@gmail.com Extended FreeBasic Neu...

I wish you luck with this.

Some decades ago, I wrote a mathematical expression parser and evaluator,
written in a version of BASIC.

To construct this, I drew up Backus-Naur diagrams.
Even large exponents were catered for.

The code must be here somewhere.