Metal quantized modifications proposal.

- Add a device param, wherever needed.
- Create new QMetal storage thing that implements QuantizedType.
- Update everywhere needed.

Fix Python.

Fixing examples.

Fix: fmt + clippy + stub.

Moving everything around.

Only missing the actual implems.

Fixing everything + adding dequantized kernels.

More work.

Fixing matmul.

Fmt + Clippy

Some clippy fixes.

Working state.

Q2K Metal -> Bugged (also present in GGML).
Q4K CPU -> Bugged (present previously, new test catch it).
Q5K CPU -> Bugged (present previously).
Q8_1 Both -> Never really implemented it seems
Q8K metal -> Never implemented in metal

Fixing Q2K bug (present in ggml).
This commit is contained in:
Nicolas Patry
2023-11-18 14:49:30 +01:00
committed by Nicolas Patry
parent 3a7304cb0d
commit f97fcd4712
32 changed files with 6408 additions and 420 deletions

View File

@ -299,21 +299,26 @@ fn main() -> Result<()> {
WhichModel::PuffinPhiV2 => Config::puffin_phi_v2(),
WhichModel::PhiHermes => Config::phi_hermes_1_3b(),
};
let (model, device) = if args.quantized {
let vb = candle_transformers::quantized_var_builder::VarBuilder::from_gguf(&filenames[0])?;
let device = candle_examples::device(args.cpu)?;
let model = if args.quantized {
let vb = candle_transformers::quantized_var_builder::VarBuilder::from_gguf(
&filenames[0],
&device,
)?;
println!("Loaded vb");
let model = match args.model {
WhichModel::V2 => QMixFormer::new_v2(&config, vb)?,
_ => QMixFormer::new(&config, vb)?,
};
(Model::Quantized(model), Device::Cpu)
println!("Loaded model");
Model::Quantized(model)
} else {
let device = candle_examples::device(args.cpu)?;
let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, DType::F32, &device)? };
let model = match args.model {
WhichModel::V2 => MixFormer::new_v2(&config, vb)?,
_ => MixFormer::new(&config, vb)?,
};
(Model::MixFormer(model), device)
Model::MixFormer(model)
};
println!("loaded the model in {:?}", start.elapsed());