libmir-metal 0.3.0

Metal inference backend for libmir
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
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132

use std::{fs, path::Path};

use models::{
    execution::DecoderExecutionContract,
    layout::{DecoderConfig, ModelLayout},
    weights::TensorCatalog,
};
use serde_json::{Map, Value, json};

use super::DenseSwiGluModel;
use crate::engine::{Array, ModelTensors, Result, Stream, lowering};

#[test]
fn executes_complete_dense_models_without_affine_group_size() -> Result<()> {
    for dtype in ["F32", "F16", "BF16"] {
        execute_model(dtype)?;
    }
    Ok(())
}

fn execute_model(dtype: &str) -> Result<()> {
    let root = std::env::temp_dir()
        .join(format!("libmir-metal-dense-model-{}-{dtype}", std::process::id()));
    fs::create_dir_all(&root)?;
    write_config(&root)?;
    write_weights(&root.join("model.safetensors"), dtype)?;

    let layout = ModelLayout::inspect(&root)?;
    let decoder = DecoderConfig::from_layout(&layout)?;
    let catalog = TensorCatalog::from_layout(&layout)?;
    let contract = DecoderExecutionContract::discover(&layout, &decoder, &catalog)?;
    let lowering = lowering::plan(&contract.semantic)?;
    let stream = Stream::new_cpu()?;
    let tensors = ModelTensors::load(&root, &stream)?;
    let model = DenseSwiGluModel::load(
        &tensors,
        &decoder,
        &contract.bindings,
        lowering.layers(),
        16,
        &stream,
    )?;
    let mut cache = model.new_cache(&stream)?;
    let logits = model.forward_decode(&Array::from_u32(&[1], &[1, 1])?, &mut cache, 0, &stream)?;

    assert_eq!(model.layer_count(), 1);
    assert_eq!(logits.shape()?, vec![1, 1, 4]);
    assert!(logits.to_vec_f32_on_stream(&stream)?.iter().all(|value| value.is_finite()));
    cache.reset()?;
    let state =
        model.forward_prefill_state(&Array::from_u32(&[1, 2], &[1, 2])?, &mut cache, 0, &stream)?;
    assert_eq!(state.shape()?, vec![1, 2, 2]);
    state.async_eval()?;
    stream.synchronize()?;
    assert_eq!(cache.cached_tokens()?, 2);
    drop(tensors);
    fs::remove_dir_all(root)?;
    Ok(())
}

fn write_config(root: &Path) -> Result<()> {
    let config = json!({
        "architectures": ["MistralForCausalLM"],
        "model_type": "mistral",
        "hidden_size": 2,
        "intermediate_size": 4,
        "num_hidden_layers": 1,
        "num_attention_heads": 1,
        "num_key_value_heads": 1,
        "head_dim": 4,
        "vocab_size": 4,
        "max_position_embeddings": 32,
        "hidden_act": "silu",
        "rms_norm_eps": 0.00001,
        "rope_theta": 10000.0,
        "tie_word_embeddings": false
    });
    fs::write(root.join("config.json"), serde_json::to_vec(&config)?)?;
    Ok(())
}

fn write_weights(path: &Path, dtype: &str) -> Result<()> {
    let specs = [
        ("model.embed_tokens.weight", vec![4, 2]),
        ("model.norm.weight", vec![2]),
        ("lm_head.weight", vec![4, 2]),
        ("model.layers.0.input_layernorm.weight", vec![2]),
        ("model.layers.0.self_attn.q_proj.weight", vec![4, 2]),
        ("model.layers.0.self_attn.k_proj.weight", vec![4, 2]),
        ("model.layers.0.self_attn.v_proj.weight", vec![4, 2]),
        ("model.layers.0.self_attn.o_proj.weight", vec![2, 4]),
        ("model.layers.0.post_attention_layernorm.weight", vec![2]),
        ("model.layers.0.mlp.gate_proj.weight", vec![4, 2]),
        ("model.layers.0.mlp.up_proj.weight", vec![4, 2]),
        ("model.layers.0.mlp.down_proj.weight", vec![2, 4]),
    ];
    let mut header = Map::new();
    let mut offset = 0_usize;
    let mut payload = Vec::new();
    let element_bytes = if dtype == "F32" {
        4
    } else {
        2
    };
    for (name, shape) in &specs {
        let elements = shape.iter().product::<usize>();
        let bytes =
            elements.checked_mul(element_bytes).ok_or(crate::engine::Error::ShapeOverflow)?;
        header.insert(
            (*name).into(),
            json!({"dtype": dtype, "shape": shape, "data_offsets": [offset, offset + bytes]}),
        );
        for _ in 0..elements {
            match dtype {
                "F32" => payload.extend_from_slice(&0.125_f32.to_le_bytes()),
                "F16" => payload.extend_from_slice(&0x3000_u16.to_le_bytes()),
                "BF16" => payload.extend_from_slice(&0x3e00_u16.to_le_bytes()),
                _ => return Err(crate::engine::Error::InvalidModel(dtype.into())),
            }
        }
        offset += bytes;
    }
    let mut header = serde_json::to_string(&Value::Object(header))?;
    while !header.len().is_multiple_of(8) {
        header.push(' ');
    }
    let mut bytes = u64::try_from(header.len())?.to_le_bytes().to_vec();
    bytes.extend_from_slice(header.as_bytes());
    bytes.extend_from_slice(&payload);
    fs::write(path, bytes)?;
    Ok(())
}