rlx-runtime 0.2.1

RLX runtime — feature-gated backends, session API, compile+execute pipeline
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

// RLX — versatile ML compiler + runtime.
// Copyright (C) 2026 Eugene Hauptmann, Nataliya Kosmyna.
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 3.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.

//! Verifies F32 set_param / input feed / output read still works after
//! AutoMixedPrecision rewrites Param + Input nodes to F16 on the CPU
//! backend's arena.
//!
//! cargo run --example cpu_mixed_precision --release \
//!   --features "cpu,blas-accelerate"

#[cfg(feature = "cpu")]
fn main() {
    use rlx_ir::*;
    use rlx_runtime::{Device, Precision, PrecisionPolicy, Session};

    let build = || {
        let mut g = Graph::new("matmul_bias_gelu");
        let x = g.input("x", Shape::new(&[2, 4], DType::F32));
        let w = g.param("w", Shape::new(&[4, 3], DType::F32));
        let b = g.param("b", Shape::new(&[3], DType::F32));
        let mm = g.matmul(x, w, Shape::new(&[2, 3], DType::F32));
        let bias = g.binary(op::BinaryOp::Add, mm, b, Shape::new(&[2, 3], DType::F32));
        let out = g.activation(op::Activation::Gelu, bias, Shape::new(&[2, 3], DType::F32));
        g.set_outputs(vec![out]);
        g
    };

    let w_data = vec![1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0];
    let b_data = vec![0.5, -0.5, 0.0];
    let x_data = vec![1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0];

    println!("policy        output");
    println!("------------  ------------------------------");
    for (name, policy) in [
        ("AlwaysF32", PrecisionPolicy::AlwaysF32),
        ("AutoMixed", PrecisionPolicy::AutoMixed),
    ] {
        let session = Session::new_with_precision(Device::Cpu, Precision::F32).with_policy(policy);
        let mut compiled = session.compile(build());
        compiled.set_param("w", &w_data);
        compiled.set_param("b", &b_data);
        let out = compiled.run(&[("x", &x_data)]);
        println!("{:<13} {:?}", name, out[0]);
        // sanity: GELU(1.5) ≈ 1.40 — both F32 and F16 paths should land near here
        assert!(
            out[0][0] > 1.3 && out[0][0] < 1.5,
            "{name}: out[0]={} not within tolerance",
            out[0][0]
        );
    }
    println!("\n✓ set_param + input feed + output read survive F32→F16 rewrite");
}

#[cfg(not(feature = "cpu"))]
fn main() {
    eprintln!("requires --features cpu");
}