forked from huggingface/candle
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.rs
241 lines (213 loc) · 7.05 KB
/
main.rs
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
#[cfg(feature = "mkl")]
extern crate intel_mkl_src;
#[cfg(feature = "accelerate")]
extern crate accelerate_src;
use anyhow::{Error as E, Result};
use clap::Parser;
use candle_transformers::models::mixtral::{Config, Model};
use candle::{DType, Device, Tensor};
use candle_examples::token_output_stream::TokenOutputStream;
use candle_nn::VarBuilder;
use candle_transformers::generation::LogitsProcessor;
use hf_hub::{api::sync::Api, Repo, RepoType};
use tokenizers::Tokenizer;
struct TextGeneration {
model: Model,
device: Device,
tokenizer: TokenOutputStream,
logits_processor: LogitsProcessor,
repeat_penalty: f32,
repeat_last_n: usize,
}
impl TextGeneration {
#[allow(clippy::too_many_arguments)]
fn new(
model: Model,
tokenizer: Tokenizer,
seed: u64,
temp: Option<f64>,
top_p: Option<f64>,
repeat_penalty: f32,
repeat_last_n: usize,
device: &Device,
) -> Self {
let logits_processor = LogitsProcessor::new(seed, temp, top_p);
Self {
model,
tokenizer: TokenOutputStream::new(tokenizer),
logits_processor,
repeat_penalty,
repeat_last_n,
device: device.clone(),
}
}
fn run(&mut self, prompt: &str, sample_len: usize) -> Result<()> {
use std::io::Write;
self.tokenizer.clear();
let mut tokens = self
.tokenizer
.tokenizer()
.encode(prompt, true)
.map_err(E::msg)?
.get_ids()
.to_vec();
for &t in tokens.iter() {
if let Some(t) = self.tokenizer.next_token(t)? {
print!("{t}")
}
}
std::io::stdout().flush()?;
let mut generated_tokens = 0usize;
let eos_token = match self.tokenizer.get_token("</s>") {
Some(token) => token,
None => anyhow::bail!("cannot find the </s> token"),
};
let start_gen = std::time::Instant::now();
for index in 0..sample_len {
let context_size = if index > 0 { 1 } else { tokens.len() };
let start_pos = tokens.len().saturating_sub(context_size);
let ctxt = &tokens[start_pos..];
let input = Tensor::new(ctxt, &self.device)?.unsqueeze(0)?;
let logits = self.model.forward(&input, start_pos)?;
let logits = logits.squeeze(0)?.squeeze(0)?.to_dtype(DType::F32)?;
let logits = if self.repeat_penalty == 1. {
logits
} else {
let start_at = tokens.len().saturating_sub(self.repeat_last_n);
candle_transformers::utils::apply_repeat_penalty(
&logits,
self.repeat_penalty,
&tokens[start_at..],
)?
};
let next_token = self.logits_processor.sample(&logits)?;
tokens.push(next_token);
generated_tokens += 1;
if next_token == eos_token {
break;
}
if let Some(t) = self.tokenizer.next_token(next_token)? {
print!("{t}");
std::io::stdout().flush()?;
}
}
let dt = start_gen.elapsed();
if let Some(rest) = self.tokenizer.decode_rest().map_err(E::msg)? {
print!("{rest}");
}
std::io::stdout().flush()?;
println!(
"\n{generated_tokens} tokens generated ({:.2} token/s)",
generated_tokens as f64 / dt.as_secs_f64(),
);
Ok(())
}
}
#[derive(Parser, Debug)]
#[command(author, version, about, long_about = None)]
struct Args {
/// Run on CPU rather than on GPU.
#[arg(long)]
cpu: bool,
/// Enable tracing (generates a trace-timestamp.json file).
#[arg(long)]
tracing: bool,
#[arg(long)]
use_flash_attn: bool,
#[arg(long)]
prompt: String,
/// The temperature used to generate samples.
#[arg(long)]
temperature: Option<f64>,
/// Nucleus sampling probability cutoff.
#[arg(long)]
top_p: Option<f64>,
/// The seed to use when generating random samples.
#[arg(long, default_value_t = 299792458)]
seed: u64,
/// The length of the sample to generate (in tokens).
#[arg(long, short = 'n', default_value_t = 100)]
sample_len: usize,
#[arg(long, default_value = "mistralai/Mixtral-8x7B-v0.1")]
model_id: String,
#[arg(long, default_value = "main")]
revision: String,
#[arg(long)]
tokenizer_file: Option<String>,
#[arg(long)]
weight_files: Option<String>,
/// Penalty to be applied for repeating tokens, 1. means no penalty.
#[arg(long, default_value_t = 1.1)]
repeat_penalty: f32,
/// The context size to consider for the repeat penalty.
#[arg(long, default_value_t = 64)]
repeat_last_n: usize,
}
fn main() -> Result<()> {
use tracing_chrome::ChromeLayerBuilder;
use tracing_subscriber::prelude::*;
let args = Args::parse();
let _guard = if args.tracing {
let (chrome_layer, guard) = ChromeLayerBuilder::new().build();
tracing_subscriber::registry().with(chrome_layer).init();
Some(guard)
} else {
None
};
println!(
"avx: {}, neon: {}, simd128: {}, f16c: {}",
candle::utils::with_avx(),
candle::utils::with_neon(),
candle::utils::with_simd128(),
candle::utils::with_f16c()
);
println!(
"temp: {:.2} repeat-penalty: {:.2} repeat-last-n: {}",
args.temperature.unwrap_or(0.),
args.repeat_penalty,
args.repeat_last_n
);
let start = std::time::Instant::now();
let api = Api::new()?;
let repo = api.repo(Repo::with_revision(
args.model_id,
RepoType::Model,
args.revision,
));
let tokenizer_filename = match args.tokenizer_file {
Some(file) => std::path::PathBuf::from(file),
None => repo.get("tokenizer.json")?,
};
let filenames = match args.weight_files {
Some(files) => files
.split(',')
.map(std::path::PathBuf::from)
.collect::<Vec<_>>(),
None => candle_examples::hub_load_safetensors(&repo, "model.safetensors.index.json")?,
};
println!("retrieved the files in {:?}", start.elapsed());
let tokenizer = Tokenizer::from_file(tokenizer_filename).map_err(E::msg)?;
let start = std::time::Instant::now();
let config = Config::v0_1_8x7b(args.use_flash_attn);
let device = candle_examples::device(args.cpu)?;
let dtype = if device.is_cuda() {
DType::BF16
} else {
DType::F32
};
let vb = unsafe { VarBuilder::from_mmaped_safetensors(&filenames, dtype, &device)? };
let model = Model::new(&config, vb)?;
println!("loaded the model in {:?}", start.elapsed());
let mut pipeline = TextGeneration::new(
model,
tokenizer,
args.seed,
args.temperature,
args.top_p,
args.repeat_penalty,
args.repeat_last_n,
&device,
);
pipeline.run(&args.prompt, args.sample_len)?;
Ok(())
}