Show retrain progress in Android

tools/windows_training_agent.py

@@ -54,12 +54,14 @@ def poll_once(args: argparse.Namespace, repo_root: Path, runtime_dir: Path, log_
     if not job_id:
         return
     log(log_path, f"Claimed retrain job {job_id}")
+    report_progress(args, job_id, "running", "claimed", 2, "Задание получено Windows-agent")
     success = False
     message = ""
     summary: dict[str, Any] = {}
     try:
-        run_retrain(job, repo_root, log_path)
+        run_retrain(args, job_id, job, repo_root, log_path)
         summary = read_json(runtime_dir / "torch_retrain_guard.json")
+        report_progress(args, job_id, "running", "uploading", 72, "Обучение завершено, загружаю артефакты")
         for name in ARTIFACT_NAMES:
             path = runtime_dir / name
             if path.is_file():
@@ -75,7 +77,7 @@ def poll_once(args: argparse.Namespace, repo_root: Path, runtime_dir: Path, log_
         api_json(args, f"/api/training/jobs/{job_id}/complete", payload)
 
 
-def run_retrain(job: dict[str, Any], repo_root: Path, log_path: Path) -> None:
+def run_retrain(args: argparse.Namespace, job_id: str, job: dict[str, Any], repo_root: Path, log_path: Path) -> None:
     script = repo_root / "tools" / "run_torch_retrain.ps1"
     if not script.is_file():
         raise RuntimeError(f"retrain script not found: {script}")
@@ -103,6 +105,8 @@ def run_retrain(job: dict[str, Any], repo_root: Path, log_path: Path) -> None:
         if value not in (None, ""):
             cmd.extend([ps_arg, str(value)])
     log(log_path, "Running retrain: " + " ".join(quote_for_log(part) for part in cmd))
+    report_progress(args, job_id, "running", "training", 8, "PyTorch retrain запущен")
+    line_count = 0
     with subprocess.Popen(
         cmd,
         cwd=str(repo_root),
@@ -114,10 +118,16 @@ def run_retrain(job: dict[str, Any], repo_root: Path, log_path: Path) -> None:
     ) as process:
         assert process.stdout is not None
         for line in process.stdout:
-            log(log_path, line.rstrip())
+            message = line.rstrip()
+            log(log_path, message)
+            line_count += 1
+            if line_count == 1 or line_count % 12 == 0:
+                progress = min(70, 8 + line_count // 3)
+                report_progress(args, job_id, "running", "training", progress, message[-220:])
         code = process.wait()
     if code != 0:
         raise RuntimeError(f"retrain failed with exit code {code}")
+    report_progress(args, job_id, "running", "guard", 70, "Guard завершён, подготавливаю артефакты")
 
 
 def upload_artifact(args: argparse.Namespace, job_id: str, path: Path, log_path: Path) -> None:
@@ -137,6 +147,30 @@ def upload_artifact(args: argparse.Namespace, job_id: str, path: Path, log_path:
                 "data_base64": base64.b64encode(data).decode("ascii"),
             }
             api_json(args, f"/api/training/jobs/{job_id}/artifacts/chunk", payload, timeout=120)
+            if index == 0 or index == total - 1 or index % 10 == 0:
+                progress = 72 + int(((index + 1) / total) * 23)
+                report_progress(args, job_id, "running", "uploading", progress, f"Загружаю {path.name}: {index + 1}/{total}")
+
+
+def report_progress(
+    args: argparse.Namespace,
+    job_id: str,
+    status: str,
+    phase: str,
+    progress_percent: int,
+    message: str,
+) -> None:
+    api_json(
+        args,
+        f"/api/training/jobs/{job_id}/progress",
+        {
+            "status": status,
+            "phase": phase,
+            "progress_percent": progress_percent,
+            "message": message,
+            "worker": worker_payload(args, Path(args.repo_root).resolve()),
+        },
+    )
 
 
 def api_json(args: argparse.Namespace, path: str, payload: dict[str, Any], timeout: int = 30) -> dict[str, Any]: