refactor generate code
This commit is contained in:
@@ -62,16 +62,10 @@ fn gelab_zero_generate() -> Result<()> {
|
||||
let i_duration = i_start.elapsed();
|
||||
println!("Time elapsed in load model is: {:?}", i_duration);
|
||||
|
||||
let i_start = Instant::now();
|
||||
let res = qwen3vl.generate(mes)?;
|
||||
let i_duration = i_start.elapsed();
|
||||
println!("generate: \n {:?}", res);
|
||||
if let Some(usage) = &res.usage {
|
||||
let num_token = usage.total_tokens;
|
||||
let duration_secs = i_duration.as_secs_f64();
|
||||
let tps = num_token as f64 / duration_secs;
|
||||
println!("Tokens per second (TPS): {:.2}", tps);
|
||||
println!("usage: \n {:?}", usage);
|
||||
}
|
||||
println!("Time elapsed in generate is: {:?}", i_duration);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user