model-rs 0.1.4

A Rust CLI tool for downloading HuggingFace models and running local LLM inference
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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285

//! GGUF (quantized model) backend using candle's pure Rust quantized support
//!
//! This module provides support for loading and running GGUF format models,
//! which offer significant memory savings through quantization.
//! Uses candle-core's quantized GGUF reader and candle-transformers' quantized
//! model implementations — no C/C++ dependencies.

use crate::error::{ModelError, Result};
use crate::local::LocalModelConfig;
use candle_core::quantized::gguf_file;
use candle_core::{Device, Tensor};
use candle_transformers::models::quantized_llama::ModelWeights as QuantizedLlama;
use std::path::Path;
use tracing::info;

/// GGUF backend for quantized model inference (pure Rust)
pub struct GgufBackend {
    context_size: usize,
    quantization: String,
    gguf_path: std::path::PathBuf,
    model: QuantizedLlama,
    device: Device,
}

impl GgufBackend {
    /// Load a GGUF model from the given path
    pub fn load(config: &LocalModelConfig, gguf_path: &Path) -> Result<Self> {
        info!("Loading GGUF model from: {}", gguf_path.display());

        let quantization = Self::detect_quantization(gguf_path)?;
        info!("Detected quantization: {}", quantization);

        let device = Device::Cpu;

        let file = std::fs::File::open(gguf_path)
            .map_err(|e| ModelError::LocalModelError(format!("Failed to open GGUF file: {}", e)))?;
        let mut reader = std::io::BufReader::new(file);

        let content = gguf_file::Content::read(&mut reader)
            .map_err(|e| ModelError::LocalModelError(format!("Failed to parse GGUF file: {}", e)))?;

        let model = QuantizedLlama::from_gguf(content, &mut reader, &device)
            .map_err(|e| ModelError::LocalModelError(format!("Failed to load quantized model: {}", e)))?;

        info!("GGUF model loaded successfully (quantization: {})", quantization);

        Ok(Self {
            gguf_path: gguf_path.to_path_buf(),
            context_size: config.max_seq_len,
            quantization,
            model,
            device,
        })
    }

    /// Check if a directory contains any .gguf files
    pub fn detect_gguf(model_path: &Path) -> Result<Option<std::path::PathBuf>> {
        if !model_path.is_dir() {
            return Ok(None);
        }
        let entries = std::fs::read_dir(model_path)
            .map_err(|e| ModelError::LocalModelError(format!("Cannot read directory: {}", e)))?;
        for entry in entries.flatten() {
            if entry.path().extension().map_or(false, |ext| ext == "gguf") {
                return Ok(Some(entry.path()));
            }
        }
        Ok(None)
    }

    /// Detect quantization format from GGUF filename
    pub fn detect_quantization(path: &Path) -> Result<String> {
        let filename = path.file_name()
            .and_then(|n| n.to_str())
            .ok_or_else(|| ModelError::LocalModelError("Invalid filename".to_string()))?;

        let filename_lower = filename.to_lowercase();

        let quant = if filename_lower.contains("q2_k") {
            "Q2_K"
        } else if filename_lower.contains("q4_k_m") {
            "Q4_K_M"
        } else if filename_lower.contains("q4_k") {
            "Q4_K"
        } else if filename_lower.contains("q5_k_m") {
            "Q5_K_M"
        } else if filename_lower.contains("q5_k") {
            "Q5_K"
        } else if filename_lower.contains("q6_k") {
            "Q6_K"
        } else if filename_lower.contains("q8_0") {
            "Q8_0"
        } else if filename_lower.contains("f16") {
            "F16"
        } else {
            "Unknown"
        };

        Ok(quant.to_string())
    }

    /// Generate text from input token IDs
    pub fn generate_text(
        &mut self,
        input_ids: &[u32],
        max_tokens: usize,
        temperature: f32,
        top_p: f32,
        top_k: Option<usize>,
        eos_token: Option<u32>,
    ) -> Result<Vec<u32>> {
        let mut generated = Vec::new();

        let prompt_tensor = Tensor::new(input_ids, &self.device)?
            .unsqueeze(0)?;
        let logits = self.model.forward(&prompt_tensor, 0)?;
        let logits = logits.to_dtype(candle_core::DType::F32)?;
        let logits_vec = logits.to_vec3::<f32>()?;
        let last_logits = &logits_vec[0][logits_vec[0].len() - 1];

        let mut next = sample_token(last_logits, temperature, top_p, top_k)?;
        generated.push(next);

        for idx in 1..max_tokens {
            if let Some(eos) = eos_token {
                if next == eos {
                    break;
                }
            }

            let tensor = Tensor::new(&[next], &self.device)?.unsqueeze(0)?;
            let logits = self.model.forward(&tensor, input_ids.len() + idx - 1)?;
            let logits = logits.to_dtype(candle_core::DType::F32)?;
            let logits_vec = logits.to_vec3::<f32>()?;
            let last_logits = &logits_vec[0][0];

            next = sample_token(last_logits, temperature, top_p, top_k)?;
            generated.push(next);
        }

        Ok(generated)
    }

    /// Generate text with streaming callback
    pub fn generate_text_stream<F>(
        &mut self,
        input_ids: &[u32],
        max_tokens: usize,
        temperature: f32,
        top_p: f32,
        top_k: Option<usize>,
        eos_token: Option<u32>,
        mut callback: F,
        tokenizer: &tokenizers::Tokenizer,
    ) -> Result<()>
    where
        F: FnMut(String) -> Result<()>,
    {
        use crate::local::tokenization::stream_piece;

        let mut started = false;

        let prompt_tensor = Tensor::new(input_ids, &self.device)?
            .unsqueeze(0)?;
        let logits = self.model.forward(&prompt_tensor, 0)?;
        let logits = logits.to_dtype(candle_core::DType::F32)?;
        let logits_vec = logits.to_vec3::<f32>()?;
        let last_logits = &logits_vec[0][logits_vec[0].len() - 1];

        let mut next = sample_token(last_logits, temperature, top_p, top_k)?;

        if let Some(piece) = stream_piece(tokenizer, next, &mut started)? {
            callback(piece)?;
        }

        for idx in 1..max_tokens {
            if let Some(eos) = eos_token {
                if next == eos {
                    break;
                }
            }

            let tensor = Tensor::new(&[next], &self.device)?.unsqueeze(0)?;
            let logits = self.model.forward(&tensor, input_ids.len() + idx - 1)?;
            let logits = logits.to_dtype(candle_core::DType::F32)?;
            let logits_vec = logits.to_vec3::<f32>()?;
            let last_logits = &logits_vec[0][0];

            next = sample_token(last_logits, temperature, top_p, top_k)?;

            if let Some(piece) = stream_piece(tokenizer, next, &mut started)? {
                callback(piece)?;
            }
        }

        Ok(())
    }

    /// Get the quantization format
    pub fn quantization(&self) -> &str {
        &self.quantization
    }

    /// Get the context size
    pub fn context_size(&self) -> usize {
        self.context_size
    }

    /// Get the GGUF file path
    pub fn path(&self) -> &Path {
        &self.gguf_path
    }
}

/// Simple sampling: temperature + top-p + top-k
fn sample_token(logits: &[f32], temperature: f32, top_p: f32, top_k: Option<usize>) -> Result<u32> {
    crate::local::sampling::do_sample(logits, temperature, top_p, top_k)
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_detect_quantization() {
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q2_k.gguf")).unwrap(),
            "Q2_K"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q4_k.gguf")).unwrap(),
            "Q4_K"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q4_k_m.gguf")).unwrap(),
            "Q4_K_M"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q5_k.gguf")).unwrap(),
            "Q5_K"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q5_k_m.gguf")).unwrap(),
            "Q5_K_M"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q6_k.gguf")).unwrap(),
            "Q6_K"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-q8_0.gguf")).unwrap(),
            "Q8_0"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model-f16.gguf")).unwrap(),
            "F16"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("MODEL-Q2_K.GGUF")).unwrap(),
            "Q2_K"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("Model-Q4_K_M.GgUF")).unwrap(),
            "Q4_K_M"
        );
    }

    #[test]
    fn test_detect_quantization_unknown() {
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model.gguf")).unwrap(),
            "Unknown"
        );
        assert_eq!(
            GgufBackend::detect_quantization(Path::new("model.bin")).unwrap(),
            "Unknown"
        );
    }

    #[test]
    fn test_detect_quantization_invalid_path() {
        assert!(GgufBackend::detect_quantization(Path::new("")).is_err());
        assert!(GgufBackend::detect_quantization(Path::new("/")).is_err());
    }
}