uqff/
main.rs

1use anyhow::Result;
2use mistralrs::{
3    PagedAttentionMetaBuilder, RequestBuilder, TextMessageRole, TextMessages, UqffTextModelBuilder,
4};
5
6#[tokio::main]
7async fn main() -> Result<()> {
8    let model = UqffTextModelBuilder::new(
9        "EricB/Phi-3.5-mini-instruct-UQFF",
10        vec!["phi3.5-mini-instruct-q8_0.uqff".into()],
11    )
12    .into_inner()
13    .with_logging()
14    .with_paged_attn(|| PagedAttentionMetaBuilder::default().build())?
15    .build()
16    .await?;
17
18    let messages = TextMessages::new()
19        .add_message(
20            TextMessageRole::System,
21            "You are an AI agent with a specialty in programming.",
22        )
23        .add_message(
24            TextMessageRole::User,
25            "Hello! How are you? Please write generic binary search function in Rust.",
26        );
27
28    let response = model.send_chat_request(messages).await?;
29
30    println!("{}", response.choices[0].message.content.as_ref().unwrap());
31    dbg!(
32        response.usage.avg_prompt_tok_per_sec,
33        response.usage.avg_compl_tok_per_sec
34    );
35
36    // Next example: Return some logprobs with the `RequestBuilder`, which enables higher configurability.
37    let request = RequestBuilder::new().return_logprobs(true).add_message(
38        TextMessageRole::User,
39        "Please write a mathematical equation where a few numbers are added.",
40    );
41
42    let response = model.send_chat_request(request).await?;
43
44    println!(
45        "Logprobs: {:?}",
46        &response.choices[0]
47            .logprobs
48            .as_ref()
49            .unwrap()
50            .content
51            .as_ref()
52            .unwrap()[0..3]
53    );
54
55    Ok(())
56}
uqff/main.rs

uqff/
main.rs
