Sync local codebase to Hugging Face

main/requirements.txt

@@ -1,12 +1,12 @@
-matplotlib==3.10.8
-numpy==2.4.1
-optuna==4.6.0
-pandas==3.0.0
-scikit_learn==1.8.0
-scipy==1.17.0
-seaborn==0.13.2
-timm==1.0.24
-torch==2.9.1+rocm6.4
-torchvision==0.24.1+rocm6.4
-tqdm==4.67.1
-umap_learn==0.5.9.post2
+matplotlib==3.8.4
+numpy==1.26.4
+optuna==4.1.1
+pandas==2.2.3
+scikit-learn==1.5.2
+scipy==1.14.1
+seaborn==0.13.1
+timm==0.9.16
+torch==2.3.1
+torchvision==0.18.1
+tqdm==4.66.5
+umap-learn==0.5.7

requirements.txt

@@ -2,15 +2,15 @@ fastapi
 uvicorn
 python-multipart
 jinja2
-matplotlib==3.10.8
-numpy==2.4.1
-optuna==4.6.0
-pandas==3.0.0
-scikit-learn
-scipy==1.17.0
-seaborn==0.13.2
-timm==1.0.24
-torch==2.9.1
-torchvision==0.24.1
-tqdm==4.67.1
-umap-learn==0.5.9.post2
+matplotlib==3.8.4
+numpy==1.26.4
+optuna==4.1.1
+pandas==2.2.3
+scikit-learn==1.5.2
+scipy==1.14.1
+seaborn==0.13.1
+timm==0.9.16
+torch==2.3.1
+torchvision==0.18.1
+tqdm==4.66.5
+umap-learn==0.5.7

webserver/app.py

@@ -35,14 +35,26 @@ os.makedirs(PREDICTIONS_DIR, exist_ok=True)
 app = FastAPI(title="Raman Fine-Tune Webserver")
 templates = Jinja2Templates(directory=TEMPLATE_DIR)
 
-if multiprocessing.current_process().name == "MainProcess":
-    JOB_MANAGER = multiprocessing.Manager()
-    JOBS = JOB_MANAGER.dict()
-else:
+# Initialize multiprocessing with fallback for constrained environments
+try:
+    if multiprocessing.current_process().name == "MainProcess":
+        JOB_MANAGER = multiprocessing.Manager()
+        JOBS = JOB_MANAGER.dict()
+    else:
+        JOB_MANAGER = None
+        JOBS = {}
+except Exception as e:
+    print(f"Warning: multiprocessing.Manager() failed: {e}. Using local dict instead.")
     JOB_MANAGER = None
     JOBS = {}
+
 JOB_PROCESSES = {}
-JOB_CONTEXT = multiprocessing.get_context("spawn")
+
+try:
+    JOB_CONTEXT = multiprocessing.get_context("spawn")
+except Exception as e:
+    print(f"Warning: spawn context not available: {e}. Using default context.")
+    JOB_CONTEXT = multiprocessing
 
 
 def _save_upload(file_obj: UploadFile, dst_path: str):
@@ -223,6 +235,25 @@ def _reap_job_process(job_id: str, process: multiprocessing.Process):
     JOB_PROCESSES.pop(job_id, None)
 
 
+@app.on_event("startup")
+async def startup_event():
+    import sys
+    print("[STARTUP] Application initializing...")
+    print(f"[STARTUP] Python version: {sys.version}")
+    print(f"[STARTUP] PyTorch: {torch.__version__}")
+    print(f"[STARTUP] CUDA available: {torch.cuda.is_available()}")
+    if torch.cuda.is_available():
+        print(f"[STARTUP] CUDA device: {torch.cuda.get_device_name(0)}")
+    print(f"[STARTUP] JOB_MANAGER: {'multiprocessing.Manager' if JOB_MANAGER else 'local dict'}")
+    print(f"[STARTUP] JOB_CONTEXT: {type(JOB_CONTEXT).__name__}")
+    print("[STARTUP] Application ready!")
+
+
+@app.get("/health")
+def health_check():
+    return {"status": "ok", "cuda": torch.cuda.is_available()}
+
+
 @app.get("/")
 def index(request: Request):
     return templates.TemplateResponse(request, "index.html", {"request": request})

webserver/train_service.py

@@ -1,5 +1,6 @@
 import json
 import os
+import sys
 import traceback
 from dataclasses import dataclass
 from datetime import datetime
@@ -16,10 +17,10 @@ from webserver.preprocess_utils import augment_small_trainset, preprocess_raman_
 # Make project root importable when the web server runs from ./webserver
 ROOT_DIR = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))
 MAIN_DIR = os.path.join(ROOT_DIR, "main")
-if ROOT_DIR not in os.sys.path:
-    os.sys.path.insert(0, ROOT_DIR)
-if MAIN_DIR not in os.sys.path:
-    os.sys.path.insert(0, MAIN_DIR)
+if ROOT_DIR not in sys.path:
+    sys.path.insert(0, ROOT_DIR)
+if MAIN_DIR not in sys.path:
+    sys.path.insert(0, MAIN_DIR)
 
 from main.Ramandataset import RamanDataset
 from main.Raman_Task import (
@@ -220,6 +221,11 @@ def predict_with_checkpoint(checkpoint_path: str, spectra: np.ndarray, wavenumbe
 
 def run_finetune_job(job_id: str, input_paths: dict, run_dir: str, config: TrainConfig, jobs: dict):
     try:
+        print(f"[JOB {job_id}] Starting fine-tune job...")
+        print(f"[JOB {job_id}] Input paths: {input_paths}")
+        print(f"[JOB {job_id}] Run directory: {run_dir}")
+        print(f"[JOB {job_id}] Config: epochs={config.epochs}, batch_size={config.batch_size}")
+        
         _update_job(
             jobs,
             job_id,
@@ -231,9 +237,11 @@ def run_finetune_job(job_id: str, input_paths: dict, run_dir: str, config: Train
             total_epochs=config.epochs,
         )
 
+        print(f"[JOB {job_id}] Loading spectral data...")
         spectral = np.load(input_paths["spectral"], allow_pickle=True)
         labels = np.load(input_paths["labels"], allow_pickle=True)
         wavenumbers = np.load(input_paths["wavenumbers"], allow_pickle=True)
+        print(f"[JOB {job_id}] Data loaded: spectral {spectral.shape}, labels {labels.shape}, wavenumbers {wavenumbers.shape}")
 
         _update_job(jobs, job_id, message="Preprocessing (crop/pad/interpolate/normalize)...", progress=5, phase="preprocessing")
         processed_x, processed_labels, target_w = preprocess_raman_dataset(