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
133
134
135
136
137

mod attention;

use self::attention::packed_attention;
use super::{
    layer::DenseSwiGluLayer,
    model::{DenseSwiGluModel, OutputProjection},
};
use crate::engine::{
    Array, DecoderCache, KvCache, Result, Stream, decode_graph, decoder::forward_packed_layers,
};

impl DenseSwiGluModel {
    pub(crate) fn forward_packed_decode(
        &self,
        token_ids: &Array,
        caches: &mut [&mut DecoderCache],
        positions: &[i32],
        stream: &Stream,
    ) -> Result<Array> {
        let hidden = self.embedding.lookup(token_ids, stream)?;
        let hidden = forward_packed_layers(&self.layers, hidden, caches, positions, stream)?;
        let hidden = self.final_norm.apply(&hidden, 1.0e-6, stream)?;
        let logits = match &self.output_projection {
            OutputProjection::TiedEmbedding => self.embedding.project(&hidden, stream)?,
            OutputProjection::Linear(output) => output.forward(&hidden, stream)?,
        };
        decode_graph::export_once(&logits, stream)?;
        Ok(logits)
    }

    pub(crate) fn forward_packed_prefill_state(
        &self,
        token_ids: &Array,
        caches: &mut [&mut DecoderCache],
        positions: &[i32],
        stream: &Stream,
    ) -> Result<Array> {
        let mut hidden = self.embedding.lookup(token_ids, stream)?;
        let shape = token_ids.shape()?;
        let position = positions.iter().copied().min().map_or(Ok(0), usize::try_from)?;
        let evaluation_step = crate::engine::decoder::prefill_evaluation_step(
            caches.len(),
            usize::try_from(shape[1])?,
            position,
            self.layers.len(),
        );
        for (index, layer) in self.layers.iter().enumerate() {
            let mut layer_caches = caches
                .iter_mut()
                .map(|cache| {
                    cache.attention_caches_mut()?.get_mut(index).ok_or_else(|| {
                        crate::engine::Error::InvalidModel(format!(
                            "missing cache for layer {index}"
                        ))
                    })
                })
                .collect::<Result<Vec<_>>>()?;
            hidden = layer.mix_packed_prefill(&hidden, &mut layer_caches, positions, stream)?;
            hidden = layer.feed_forward_packed(&hidden, caches.len(), stream)?;
            if evaluation_step
                .is_some_and(|step| (index + 1) % step == 0 || index + 1 == self.layers.len())
            {
                hidden.async_eval()?;
                stream.synchronize()?;
            }
        }
        Ok(hidden)
    }
}

impl DenseSwiGluLayer {
    pub(super) fn mix_packed(
        &self,
        input: &Array,
        caches: &mut [&mut KvCache],
        positions: &[i32],
        stream: &Stream,
    ) -> Result<Array> {
        let normalized = self.weights.input_norm.apply(input, self.config.rms_norm_eps, stream)?;
        let attention = packed_attention(
            &normalized,
            &self.weights.attention,
            self.fused_attention.as_ref(),
            self.config,
            caches,
            positions,
            false,
            stream,
        )?;
        input.add(&attention, stream)
    }

    pub(super) fn mix_packed_prefill(
        &self,
        input: &Array,
        caches: &mut [&mut KvCache],
        positions: &[i32],
        stream: &Stream,
    ) -> Result<Array> {
        let normalized = self.weights.input_norm.apply(input, self.config.rms_norm_eps, stream)?;
        let attention = packed_attention(
            &normalized,
            &self.weights.attention,
            self.fused_attention.as_ref(),
            self.config,
            caches,
            positions,
            true,
            stream,
        )?;
        input.add(&attention, stream)
    }

    pub(super) fn feed_forward_packed(
        &self,
        input: &Array,
        batch_size: usize,
        stream: &Stream,
    ) -> Result<Array> {
        let normalized =
            self.weights
                .post_attention_norm
                .apply(input, self.config.rms_norm_eps, stream)?;
        let fused = (batch_size <= 2).then_some(self.fused_gate_up.as_ref()).flatten();
        let (gate, up) = fused.map_or_else(
            || {
                Ok((
                    self.weights.mlp.gate.forward(&normalized, stream)?,
                    self.weights.mlp.up.forward(&normalized, stream)?,
                ))
            },
            |fused| fused.forward_pair(&normalized, stream),
        )?;
        let activated = gate.silu_mul(&up, stream)?;
        input.add(&self.weights.mlp.down.forward(&activated, stream)?, stream)
    }
}