Files
ml/hanzo-nn/examples/basic_optimizer.rs
zeekay 75125f041f Update to candle 0.9.2-alpha.2 with full hanzo naming and latest upstream changes
- Updated all crate versions to 0.9.2-alpha.2
- Synced source code with latest upstream candle changes
- Added hanzo-ug crate as wrapper for ug functionality
- Fixed all import statements to use hanzo-* naming
- Updated dependencies to latest versions:
  * cudarc: 0.16.3 -> 0.18.2
  * gemm: 0.17.0 -> 0.18.2
  * safetensors: 0.4.1 -> 0.6.0
  * ug: 0.4.0 -> 0.5.0
  * criterion: 0.5.1 -> 0.7.0
- Added float8 support for enhanced precision
- Updated Metal backend to use objc2-metal/objc2-foundation
- All core crates now compile and build successfully

Core working crates:
- hanzo-ml (main ML framework)
- hanzo-nn (neural network layers)
- hanzo-transformers (transformer models)
- hanzo-datasets (data loading utilities)
- hanzo-ug (unified GPU backend)
- hanzo-kernels (CUDA kernels)
- hanzo-metal-kernels (Metal kernels)
- hanzo-flash-attn (flash attention)
- hanzo-onnx (ONNX support)
2026-01-16 17:17:01 -08:00

40 lines
1.3 KiB
Rust

#[cfg(feature = "mkl")]
extern crate intel_mkl_src;
#[cfg(feature = "accelerate")]
extern crate accelerate_src;
use hanzo_ml::{DType, Device, Result, Tensor};
use hanzo_nn::{linear, AdamW, Linear, Module, Optimizer, ParamsAdamW, VarBuilder, VarMap};
fn gen_data() -> Result<(Tensor, Tensor)> {
// Generate some sample linear data.
let w_gen = Tensor::new(&[[3f32, 1.]], &Device::Cpu)?;
let b_gen = Tensor::new(-2f32, &Device::Cpu)?;
let gen = Linear::new(w_gen, Some(b_gen));
let sample_xs = Tensor::new(&[[2f32, 1.], [7., 4.], [-4., 12.], [5., 8.]], &Device::Cpu)?;
let sample_ys = gen.forward(&sample_xs)?;
Ok((sample_xs, sample_ys))
}
fn main() -> Result<()> {
let (sample_xs, sample_ys) = gen_data()?;
// Use backprop to run a linear regression between samples and get the coefficients back.
let varmap = VarMap::new();
let vb = VarBuilder::from_varmap(&varmap, DType::F32, &Device::Cpu);
let model = linear(2, 1, vb.pp("linear"))?;
let params = ParamsAdamW {
lr: 0.1,
..Default::default()
};
let mut opt = AdamW::new(varmap.all_vars(), params)?;
for step in 0..10000 {
let ys = model.forward(&sample_xs)?;
let loss = ys.sub(&sample_ys)?.sqr()?.sum_all()?;
opt.backward_step(&loss)?;
println!("{step} {}", loss.to_vec0::<f32>()?);
}
Ok(())
}