llama-rs 0.17.0

A high-performance Rust implementation of llama.cpp - LLM inference engine with full GGUF support
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
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
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196

//! Validate that our inference produces sensible intermediate values.

use llama_rs::backend::Backend;
use llama_rs::backend::cpu::CpuBackend;
use llama_rs::gguf::GgufFile;
use llama_rs::tensor::{DType, Tensor};
use std::path::Path;

fn load_tensor(gguf: &GgufFile, name: &str) -> Tensor {
    let tensor_info = gguf.data.get_tensor(name).unwrap();
    let tensor_data = gguf.tensor_data(name).unwrap();
    let shape: Vec<usize> = tensor_info.dims.iter().map(|&d| d as usize).collect();
    let dtype = DType::from(tensor_info.dtype);
    Tensor::new(tensor_data.to_vec(), shape, dtype).unwrap()
}

fn dequant(backend: &CpuBackend, t: &Tensor) -> Vec<f32> {
    let numel = t.numel();
    let mut out = Tensor::zeros(vec![numel], DType::F32);
    backend.dequantize(t, &mut out).unwrap();
    out.as_f32().unwrap().to_vec()
}

fn stats(x: &[f32]) -> (f32, f32, f32, f32) {
    let min = x.iter().cloned().fold(f32::INFINITY, f32::min);
    let max = x.iter().cloned().fold(f32::NEG_INFINITY, f32::max);
    let mean = x.iter().sum::<f32>() / x.len() as f32;
    let variance: f32 = x.iter().map(|v| (v - mean).powi(2)).sum::<f32>() / x.len() as f32;
    let std = variance.sqrt();
    (min, max, mean, std)
}

fn main() {
    let model_path = "/home/joseph/Models/qwen2.5-0.5b-instruct-q4_k_m.gguf";
    let gguf = GgufFile::open(Path::new(model_path)).expect("Failed to open GGUF");
    let backend = CpuBackend::new();

    println!("=== Validating Embedding ===");

    let emb_tensor = load_tensor(&gguf, "token_embd.weight");
    println!(
        "Embedding tensor shape: {:?}, dtype: {:?}",
        emb_tensor.shape(),
        emb_tensor.dtype()
    );

    let emb = dequant(&backend, &emb_tensor);
    let (e_min, e_max, e_mean, e_std) = stats(&emb);
    println!(
        "Embedding stats: min={:.4}, max={:.4}, mean={:.4}, std={:.4}",
        e_min, e_max, e_mean, e_std
    );

    // Check a few specific embeddings
    let hidden_size = 896;
    let vocab_size = 151936;

    println!();
    println!("Sample embeddings:");
    for tok in [0, 1, 16, 17, 28, 100, 1000] {
        let start = tok * hidden_size;
        let end = start + hidden_size;
        let tok_emb = &emb[start..end];
        let (t_min, t_max, t_mean, t_std) = stats(tok_emb);
        println!(
            "  Token {:5}: min={:+.4}, max={:+.4}, mean={:+.4}, std={:.4}, first3={:.4?}",
            tok,
            t_min,
            t_max,
            t_mean,
            t_std,
            &tok_emb[..3]
        );
    }

    // Verify that different tokens have different embeddings
    println!();
    println!("Verifying token embeddings are distinct:");
    let emb_16 = &emb[16 * hidden_size..(16 + 1) * hidden_size];
    let emb_17 = &emb[17 * hidden_size..(17 + 1) * hidden_size];
    let emb_28 = &emb[28 * hidden_size..(28 + 1) * hidden_size];

    let diff_16_17: f32 = emb_16
        .iter()
        .zip(emb_17.iter())
        .map(|(a, b)| (a - b).abs())
        .sum();
    let diff_16_28: f32 = emb_16
        .iter()
        .zip(emb_28.iter())
        .map(|(a, b)| (a - b).abs())
        .sum();
    let diff_17_28: f32 = emb_17
        .iter()
        .zip(emb_28.iter())
        .map(|(a, b)| (a - b).abs())
        .sum();

    println!("  |emb[16] - emb[17]| = {:.4} (should be > 0)", diff_16_17);
    println!("  |emb[16] - emb[28]| = {:.4} (should be > 0)", diff_16_28);
    println!("  |emb[17] - emb[28]| = {:.4} (should be > 0)", diff_17_28);

    // Check output projection
    println!();
    println!("=== Validating Output Projection ===");

    let out_tensor = load_tensor(&gguf, "output.weight");
    println!(
        "Output weight shape: {:?}, dtype: {:?}",
        out_tensor.shape(),
        out_tensor.dtype()
    );

    let out_w = dequant(&backend, &out_tensor);
    let (o_min, o_max, o_mean, o_std) = stats(&out_w);
    println!(
        "Output weight stats: min={:.4}, max={:.4}, mean={:.4}, std={:.4}",
        o_min, o_max, o_mean, o_std
    );

    // Check specific rows (for tokens 16, 17, 28)
    // For GGUF shape [896, 151936], row j (for output token j) is at data[j*896..(j+1)*896]
    println!();
    println!("Sample output rows (for computing logits):");
    for tok in [16, 17, 28, 0, 1] {
        let start = tok * hidden_size;
        let end = start + hidden_size;
        let row = &out_w[start..end];
        let (r_min, r_max, r_mean, r_std) = stats(row);
        println!(
            "  Token {:5}: min={:+.4}, max={:+.4}, mean={:+.4}, std={:.4}",
            tok, r_min, r_max, r_mean, r_std
        );
    }

    // Verify output weights vary between tokens
    let row_16 = &out_w[16 * hidden_size..17 * hidden_size];
    let row_17 = &out_w[17 * hidden_size..18 * hidden_size];
    let diff_rows: f32 = row_16
        .iter()
        .zip(row_17.iter())
        .map(|(a, b)| (a - b).abs())
        .sum();
    println!();
    println!(
        "  |out_row[16] - out_row[17]| = {:.4} (should be > 0)",
        diff_rows
    );

    // Manual logit computation test
    println!();
    println!("=== Manual Logit Test ===");

    // Use a random-ish hidden state
    let test_hidden: Vec<f32> = (0..hidden_size)
        .map(|i| ((i as f32 * 0.01).sin()))
        .collect();
    let (h_min, h_max, h_mean, _) = stats(&test_hidden);
    println!(
        "Test hidden: min={:.4}, max={:.4}, mean={:.4}",
        h_min, h_max, h_mean
    );

    // Compute logit for token 17
    let logit_17: f32 = test_hidden
        .iter()
        .zip(row_17.iter())
        .map(|(h, w)| h * w)
        .sum();
    println!("Logit for token 17: {:.4}", logit_17);

    // Compute all logits
    let mut logits = vec![0.0f32; vocab_size];
    for j in 0..vocab_size {
        let row_start = j * hidden_size;
        let mut sum = 0.0f32;
        for i in 0..hidden_size {
            sum += test_hidden[i] * out_w[row_start + i];
        }
        logits[j] = sum;
    }

    let (l_min, l_max, l_mean, l_std) = stats(&logits);
    println!(
        "All logits: min={:.4}, max={:.4}, mean={:.4}, std={:.4}",
        l_min, l_max, l_mean, l_std
    );

    // Find top predictions
    let mut indexed: Vec<(usize, f32)> = logits.iter().cloned().enumerate().collect();
    indexed.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap());
    println!("Top 5 predictions:");
    for i in 0..5 {
        println!("  {}: token {} = {:.4}", i + 1, indexed[i].0, indexed[i].1);
    }
}