File size: 6,976 Bytes
54a950a |
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 |
use crossbeam_channel::{bounded, Sender, Receiver};
use dashmap::DashMap;
use regex::Regex;
use reqwest::Client;
use scraper::{Html, Selector};
use std::io::{self, Write, BufWriter};
use std::sync::atomic::{AtomicU64, Ordering};
use std::sync::Arc;
use std::time::{Duration, Instant};
use std::hash::{Hash, Hasher};
use std::collections::hash_map::DefaultHasher;
use tokenizers::Tokenizer;
use url::Url;
lazy_static::lazy_static! {
static ref MULTI_SPACE: Regex = Regex::new(r"\s+").unwrap();
}
struct Feeder {
client: Client,
visited: DashMap<u64, ()>,
url_queue: Sender<String>,
url_recv: Receiver<String>,
token_tx: Sender<u32>,
tokenizer: Tokenizer,
tokens_out: AtomicU64,
pages: AtomicU64,
start: Instant,
}
fn hash_url(url: &str) -> u64 {
let mut h = DefaultHasher::new();
url.hash(&mut h);
h.finish()
}
impl Feeder {
fn new(token_tx: Sender<u32>, queue_size: usize) -> Self {
let (url_tx, url_rx) = bounded(queue_size);
// Load tokenizer from file (pre-downloaded)
let tokenizer = Tokenizer::from_file("/workspace/tokenizer.json")
.expect("Failed to load tokenizer - download first with: curl -L 'https://huggingface.co/gpt2/resolve/main/tokenizer.json' -o /workspace/tokenizer.json");
Self {
client: Client::builder()
.timeout(Duration::from_secs(8))
.pool_max_idle_per_host(100)
.gzip(true).brotli(true).deflate(true)
.redirect(reqwest::redirect::Policy::limited(3))
.danger_accept_invalid_certs(true)
.user_agent("Mozilla/5.0 (Windows NT 10.0; Win64; x64) Chrome/120.0.0.0")
.build().unwrap(),
visited: DashMap::new(),
url_queue: url_tx,
url_recv: url_rx,
token_tx,
tokenizer,
tokens_out: AtomicU64::new(0),
pages: AtomicU64::new(0),
start: Instant::now(),
}
}
fn extract_text(html: &str) -> String {
let doc = Html::parse_document(html);
let sel = Selector::parse("p,h1,h2,h3,h4,li,td,article,span").unwrap();
let mut text = String::with_capacity(html.len() / 4);
for el in doc.select(&sel) {
let t: String = el.text().collect();
if t.len() > 20 {
text.push_str(t.trim());
text.push(' ');
}
}
MULTI_SPACE.replace_all(&text, " ").trim().to_string()
}
fn extract_links(html: &str, base: &Url) -> Vec<String> {
let doc = Html::parse_document(html);
let sel = Selector::parse("a[href]").unwrap();
let mut links = Vec::with_capacity(30);
for el in doc.select(&sel) {
if let Some(href) = el.value().attr("href") {
if let Ok(u) = base.join(href) {
let s = u.to_string();
if s.starts_with("http") && !s.contains("login")
&& !s.contains(".pdf") && !s.contains(".jpg") {
links.push(s);
}
}
}
}
links
}
async fn crawl_and_feed(&self, url: String) {
let h = hash_url(&url);
if self.visited.contains_key(&h) { return; }
self.visited.insert(h, ());
// Fetch
let resp = match self.client.get(&url).send().await {
Ok(r) if r.status().is_success() => r,
_ => return,
};
let ct = resp.headers().get("content-type")
.and_then(|v| v.to_str().ok()).unwrap_or("");
if !ct.contains("text/html") { return; }
let html = match resp.text().await { Ok(t) => t, _ => return };
let base = match Url::parse(&url) { Ok(u) => u, _ => return };
// Extract and queue links
let links = Self::extract_links(&html, &base);
for link in links.into_iter().take(15) {
let _ = self.url_queue.try_send(link);
}
// Extract text
let text = Self::extract_text(&html);
if text.len() < 100 { return; }
// Tokenize and send
if let Ok(encoding) = self.tokenizer.encode(text, false) {
for &id in encoding.get_ids() {
if self.token_tx.send(id).is_err() {
return; // Consumer closed
}
self.tokens_out.fetch_add(1, Ordering::Relaxed);
}
}
self.pages.fetch_add(1, Ordering::Relaxed);
}
fn print_stats(&self) {
let e = self.start.elapsed().as_secs_f64();
let tok = self.tokens_out.load(Ordering::Relaxed);
let pgs = self.pages.load(Ordering::Relaxed);
eprintln!("[FEEDER {:.0}s] {} pages | {} tok | {:.0} tok/s",
e, pgs, tok, tok as f64 / e);
}
}
#[tokio::main(flavor = "multi_thread", worker_threads = 64)]
async fn main() {
eprintln!("Wire-Speed Feeder starting...");
// Channel to Python trainer
let (token_tx, token_rx) = bounded::<u32>(100_000);
let feeder = Arc::new(Feeder::new(token_tx, 500_000));
// Seed URLs
let seeds = vec![
"https://en.wikipedia.org/wiki/Main_Page",
"https://news.ycombinator.com/",
"https://reddit.com/r/all",
"https://bbc.com/news",
"https://reuters.com",
"https://arxiv.org",
"https://stackoverflow.com/questions",
"https://medium.com",
"https://theguardian.com",
"https://nature.com",
"https://github.com/explore",
"https://nytimes.com",
];
for seed in seeds {
let _ = feeder.url_queue.try_send(seed.to_string());
}
// Stats printer
let stats_feeder = Arc::clone(&feeder);
tokio::spawn(async move {
loop {
tokio::time::sleep(Duration::from_secs(10)).await;
stats_feeder.print_stats();
}
});
// Output thread - writes tokens to stdout
std::thread::spawn(move || {
let stdout = io::stdout();
let mut out = BufWriter::with_capacity(1 << 16, stdout.lock());
while let Ok(token) = token_rx.recv() {
let _ = writeln!(out, "{}", token);
}
});
// Worker pool - 500 concurrent fetchers
let mut handles = Vec::with_capacity(500);
for _ in 0..500 {
let f = Arc::clone(&feeder);
handles.push(tokio::spawn(async move {
loop {
if let Ok(url) = f.url_recv.try_recv() {
f.crawl_and_feed(url).await;
} else {
tokio::time::sleep(Duration::from_micros(100)).await;
}
}
}));
}
// Run forever
loop {
tokio::time::sleep(Duration::from_secs(3600)).await;
}
}
|