mirror of
https://github.com/huggingface/candle.git
synced 2025-06-16 10:38:54 +00:00
157 lines
4.0 KiB
Rust
157 lines
4.0 KiB
Rust
#[cfg(feature = "mkl")]
|
|
extern crate intel_mkl_src;
|
|
|
|
#[cfg(feature = "accelerate")]
|
|
extern crate accelerate_src;
|
|
|
|
use anyhow::Result;
|
|
use clap::Parser;
|
|
|
|
use candle_transformers::models::csm::{Config, Model};
|
|
|
|
use candle::DType;
|
|
use candle_nn::VarBuilder;
|
|
use hf_hub::{api::sync::Api, Repo, RepoType};
|
|
|
|
#[derive(Clone, Debug, Copy, PartialEq, Eq, clap::ValueEnum)]
|
|
enum Which {
|
|
#[value(name = "1b")]
|
|
Csm1b,
|
|
}
|
|
|
|
#[derive(Parser, Debug)]
|
|
#[command(author, version, about, long_about = None)]
|
|
struct Args {
|
|
/// Run on CPU rather than on GPU.
|
|
#[arg(long)]
|
|
cpu: bool,
|
|
|
|
/// Enable tracing (generates a trace-timestamp.json file).
|
|
#[arg(long)]
|
|
tracing: bool,
|
|
|
|
#[arg(long)]
|
|
use_flash_attn: bool,
|
|
|
|
#[arg(long)]
|
|
prompt: String,
|
|
|
|
/// The temperature used to generate samples.
|
|
#[arg(long, default_value_t = 0.7)]
|
|
temperature: f64,
|
|
|
|
/// Nucleus sampling probability cutoff.
|
|
#[arg(long)]
|
|
top_p: Option<f64>,
|
|
|
|
/// Only sample among the top K samples.
|
|
#[arg(long)]
|
|
top_k: Option<usize>,
|
|
|
|
/// The seed to use when generating random samples.
|
|
#[arg(long, default_value_t = 299792458)]
|
|
seed: u64,
|
|
|
|
/// The length of the sample to generate (in tokens).
|
|
#[arg(long, short = 'n', default_value_t = 10000)]
|
|
sample_len: usize,
|
|
|
|
/// The model size to use.
|
|
#[arg(long, default_value = "1b")]
|
|
which: Which,
|
|
|
|
#[arg(long)]
|
|
model_id: Option<String>,
|
|
|
|
#[arg(long, default_value = "main")]
|
|
revision: String,
|
|
|
|
#[arg(long)]
|
|
tokenizer: Option<String>,
|
|
|
|
#[arg(long)]
|
|
config: Option<String>,
|
|
|
|
#[arg(long)]
|
|
weights: Option<String>,
|
|
|
|
/// Penalty to be applied for repeating tokens, 1. means no penalty.
|
|
#[arg(long, default_value_t = 1.1)]
|
|
repeat_penalty: f32,
|
|
|
|
/// The context size to consider for the repeat penalty.
|
|
#[arg(long, default_value_t = 64)]
|
|
repeat_last_n: usize,
|
|
}
|
|
|
|
fn main() -> Result<()> {
|
|
use tracing_chrome::ChromeLayerBuilder;
|
|
use tracing_subscriber::prelude::*;
|
|
|
|
let args = Args::parse();
|
|
|
|
let _guard = if args.tracing {
|
|
let (chrome_layer, guard) = ChromeLayerBuilder::new().build();
|
|
tracing_subscriber::registry().with(chrome_layer).init();
|
|
Some(guard)
|
|
} else {
|
|
None
|
|
};
|
|
println!(
|
|
"avx: {}, neon: {}, simd128: {}, f16c: {}",
|
|
candle::utils::with_avx(),
|
|
candle::utils::with_neon(),
|
|
candle::utils::with_simd128(),
|
|
candle::utils::with_f16c()
|
|
);
|
|
println!(
|
|
"temp: {:.2} repeat-penalty: {:.2} repeat-last-n: {}",
|
|
args.temperature, args.repeat_penalty, args.repeat_last_n
|
|
);
|
|
|
|
let start = std::time::Instant::now();
|
|
let api = Api::new()?;
|
|
let model_id = match args.model_id {
|
|
Some(model_id) => model_id,
|
|
None => {
|
|
let name = match args.which {
|
|
Which::Csm1b => "sesame/csm-1b",
|
|
};
|
|
name.to_string()
|
|
}
|
|
};
|
|
let repo = api.repo(Repo::with_revision(
|
|
model_id,
|
|
RepoType::Model,
|
|
args.revision,
|
|
));
|
|
let filenames = match args.weights {
|
|
Some(files) => files
|
|
.split(',')
|
|
.map(std::path::PathBuf::from)
|
|
.collect::<Vec<_>>(),
|
|
None => vec![repo.get("model.safetensors")?],
|
|
};
|
|
println!("retrieved the files in {:?}", start.elapsed());
|
|
|
|
let start = std::time::Instant::now();
|
|
let config: Config = match args.config {
|
|
Some(config_file) => serde_json::from_slice(&std::fs::read(config_file)?)?,
|
|
None => {
|
|
let config_file = repo.get("config.json")?;
|
|
serde_json::from_slice(&std::fs::read(config_file)?)?
|
|
}
|
|
};
|
|
let device = candle_examples::device(args.cpu)?;
|
|
let (_model, _device) = {
|
|
let dtype = DType::F32;
|
|
let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
|
|
let model = Model::new(&config, vb)?;
|
|
(model, device)
|
|
};
|
|
|
|
println!("loaded the model in {:?}", start.elapsed());
|
|
|
|
Ok(())
|
|
}
|