feat(server): improve download logging (#66)

launcher/src/main.rs

@@ -22,8 +22,8 @@ struct Args {
     model_id: String,
     #[clap(long, env)]
     revision: Option<String>,
-    #[clap(long, env)]
+    #[clap(default_value = "1", long, env)]
-    num_shard: Option<usize>,
+    num_shard: usize,
     #[clap(long, env)]
     quantize: bool,
     #[clap(default_value = "128", long, env)]
@@ -54,6 +54,16 @@ struct Args {
 
 fn main() -> ExitCode {
     // Pattern match configuration
+    let args = Args::parse();
+
+    if args.json_output {
+        tracing_subscriber::fmt().json().init();
+    } else {
+        tracing_subscriber::fmt().compact().init();
+    }
+
+    tracing::info!("{:?}", args);
+
     let Args {
         model_id,
         revision,
@@ -71,16 +81,7 @@ fn main() -> ExitCode {
         weights_cache_override,
         json_output,
         otlp_endpoint,
-    } = Args::parse();
+    } = args;
-
-    if json_output {
-        tracing_subscriber::fmt().json().init();
-    } else {
-        tracing_subscriber::fmt().compact().init();
-    }
-
-    // By default we only have one master shard
-    let num_shard = num_shard.unwrap_or(1);
 
     // Signal handler
     let running = Arc::new(AtomicBool::new(true));
@@ -123,7 +124,7 @@ fn main() -> ExitCode {
         };
 
         // Start process
-        tracing::info!("Starting download");
+        tracing::info!("Starting download process.");
         let mut download_process = match Popen::create(
             &download_argv,
             PopenConfig {
@@ -184,7 +185,7 @@ fn main() -> ExitCode {
                         }
                     }
                     _ => {
-                        tracing::error!("Download process exited with an unkown status.");
+                        tracing::error!("Download process exited with an unknown status.");
                         return ExitCode::FAILURE;
                     }
                 }

server/text_generation/utils/convert.py

@@ -83,14 +83,10 @@ def convert_files(pt_files: List[Path], st_files: List[Path]):
     ]
 
     # We do this instead of using tqdm because we want to parse the logs with the launcher
-    logger.info("Converting weights...")
     start_time = time.time()
     for i, future in enumerate(concurrent.futures.as_completed(futures)):
         elapsed = timedelta(seconds=int(time.time() - start_time))
         remaining = len(futures) - (i + 1)
-        if remaining != 0:
+        eta = (elapsed / (i + 1)) * remaining if remaining > 0 else 0
-            eta = (elapsed / (i + 1)) * remaining
-        else:
-            eta = 0
 
         logger.info(f"Convert: [{i + 1}/{len(futures)}] -- ETA: {eta}")

server/text_generation/utils/hub.py

@@ -134,6 +134,7 @@ def download_weights(
             logger.info(f"File {filename} already present in cache.")
             return local_file
 
+        logger.info(f"Starting {filename} download.")
         start_time = time.time()
         local_file = hf_hub_download(
             filename=filename,
@@ -144,7 +145,7 @@ def download_weights(
         logger.info(
             f"Downloaded {filename} at {local_file} in {timedelta(seconds=int(time.time() - start_time))}."
         )
-        return local_file
+        return Path(local_file)
 
     executor = ThreadPoolExecutor(max_workers=5)
     futures = [
@@ -152,18 +153,14 @@ def download_weights(
     ]
 
     # We do this instead of using tqdm because we want to parse the logs with the launcher
-    logger.info("Downloading weights...")
     start_time = time.time()
     files = []
     for i, future in enumerate(concurrent.futures.as_completed(futures)):
         elapsed = timedelta(seconds=int(time.time() - start_time))
         remaining = len(futures) - (i + 1)
-        if remaining != 0:
+        eta = (elapsed / (i + 1)) * remaining if remaining > 0 else 0
-            eta = (elapsed / (i + 1)) * remaining
-        else:
-            eta = 0
 
         logger.info(f"Download: [{i + 1}/{len(futures)}] -- ETA: {eta}")
-        files.append(Path(future.result()))
+        files.append(future.result())
 
-    return [Path(p) for p in files]
+    return files