add qwen2.5vl model

This commit is contained in:
jhqxxx
2025-09-22 16:38:12 +08:00
parent 17a2540484
commit 0b0076cdc9
25 changed files with 6662 additions and 0 deletions
+56
View File
@@ -0,0 +1,56 @@
use std::time::Instant;
use aha::{models::{qwen2_5vl::generate::Qwen2_5VLGenerateModel, GenerateModel}, ModelType};
use anyhow::{Result};
use candle_core::{DType, Device};
use openai_dive::v1::resources::chat::ChatCompletionParameters;
#[test]
fn qwen2_5vl_generate() -> Result<()> {
// test with cpu :(太慢了, : RUST_BACKTRACE=1 cargo test qwen2_5vl_generate -- --nocapture
// test with cuda: RUST_BACKTRACE=1 cargo test -F cuda qwen2_5vl_generate -- --nocapture
// test with cuda+flash-attn: RUST_BACKTRACE=1 cargo test -F cuda,flash-attn qwen2_5vl_generate -- --nocapture
let device = Device::cuda_if_available(0)?;
let dtype = DType::BF16;
let model_path = "/home/jhq/huggingface_model/Qwen/Qwen2.5-VL-3B-Instruct/";
let message = r#"
{
"model": "qwen2.5vl",
"messages": [
{
"role": "user",
"content": [
{
"type": "image",
"image_url":
{
"url": "file://./assets/img/ocr_test.png"
}
},
{
"type": "text",
"text": "请分析图片并提取所有可见文本内容,按从左到右、从上到下的布局,返回纯文本"
}
]
}
]
}
"#;
let mes:ChatCompletionParameters = serde_json::from_str(message)?;
let i_start = Instant::now();
// let mut model = Qwen2_5VLGenerateModel::init(model_path, &device, dtype)?;
let mut model = ModelType::init(ModelType::Qwen2_5VL, model_path, None, None)?;
let i_duration = i_start.elapsed();
println!("Time elapsed in load model is: {:?}", i_duration);
let i_start = Instant::now();
let result = model.generate(mes)?;
println!("generate: \n{}", result);
let i_duration = i_start.elapsed();
println!("Time elapsed in generate is: {:?}", i_duration);
Ok(())
}