debug msg hidden, eval limit

app.py

@@ -5,15 +5,18 @@ os.makedirs(os.environ['HF_HOME'], exist_ok=True)  # Ensure directory exists
 
 from fastapi import FastAPI
 from fastapi.responses import HTMLResponse
-from qwen_classifier.predict import predict_single  # Your existing function
-from qwen_classifier.evaluate import evaluate_batch  # Your existing function
-from qwen_classifier.globals import model, tokenizer
 import torch
 from transformers import AutoTokenizer
 from huggingface_hub import login
+from pydantic import BaseModel
+import warnings
+from transformers import logging as hf_logging
+
+from qwen_classifier.predict import predict_single  # Your existing function
+from qwen_classifier.evaluate import evaluate_batch  # Your existing function
+from qwen_classifier.globals import global_model, global_tokenizer
 from qwen_classifier.model import QwenClassifier
 from qwen_classifier.config import HF_REPO
-from pydantic import BaseModel
 
 
 app = FastAPI(title="Qwen Classifier")
@@ -21,6 +24,14 @@ hf_repo = os.getenv("HF_REPO")
 if not hf_repo:
     hf_repo = HF_REPO
 
+debug = False
+if not debug:
+    warnings.filterwarnings("ignore", message="Some weights of the model checkpoint")
+    hf_logging.set_verbosity_error()
+else:
+    hf_logging.set_verbosity_info()
+    warnings.simplefilter("default")
+
 # Add this endpoint
 @app.get("/", response_class=HTMLResponse)
 def home():
@@ -44,7 +55,7 @@ def home():
 
 @app.on_event("startup")
 async def load_model():
-    global model, tokenizer
+    global global_model, global_tokenizer
     # Warm up GPU
     torch.zeros(1).cuda() 
     # Read HF_TOKEN from Hugging Face Space secrets
@@ -60,7 +71,7 @@ async def load_model():
     model = QwenClassifier.from_pretrained(
         hf_repo,
     )
-    tokenizer = AutoTokenizer.from_pretrained(hf_repo)
+    global_tokenizer = AutoTokenizer.from_pretrained(hf_repo)
     print("Model loaded successfully!")
 
 

qwen_classifier/evaluate.py

@@ -9,8 +9,9 @@ import torch
 from datasets import Dataset
 from torch.utils.data import DataLoader
 import requests
-from .config import TAG_NAMES, DEVICE, SPACE_URL
-from .globals import model, tokenizer
+
+from .config import TAG_NAMES, DEVICE, SPACE_URL, EVAL_LIMIT
+from .globals import global_model, global_tokenizer
 
 def load_data(test_data_path):
     # zip file handler
@@ -22,7 +23,7 @@ def load_data(test_data_path):
     features = ["prob_desc_description","prob_desc_input_spec","prob_desc_output_spec"]
     cols = features + ["tags"]
     # extract a specific file from the zip container
-    for name in names[1:]:
+    for name in names[1:1+EVAL_LIMIT]:
         f = zip_file.open(name)
 
     # save the extraced file
@@ -73,15 +74,15 @@ def evaluate_batch(file_path, hf_repo, backend="local", hf_token=None):
         raise ValueError(f"Unknown backend: {backend}")
 
 def _evaluate_local(test_data_path, hf_repo):
-    global model, tokenizer
+    global global_model, global_tokenizer
     
     # Lazy-loading to avoid slow startup
-    if model is None:
+    if global_model is None:
         from .model import QwenClassifier
         from transformers import AutoTokenizer
         
-        model = QwenClassifier.from_pretrained(hf_repo).eval()
-        tokenizer = AutoTokenizer.from_pretrained(hf_repo)
+        global_model = QwenClassifier.from_pretrained(hf_repo).eval()
+        global_tokenizer = AutoTokenizer.from_pretrained(hf_repo)
     df = load_data(test_data_path)
     df = preprocessing(df)
 
@@ -89,7 +90,7 @@ def _evaluate_local(test_data_path, hf_repo):
 
     # Then apply tokenization
     def tokenize_function(examples):
-        return tokenizer(examples["text"], padding="max_length", truncation=True, max_length=512)
+        return global_tokenizer(examples["text"], padding="max_length", truncation=True, max_length=512)
 
     dataset = hf_dataset.map(tokenize_function, batched=True)
 
@@ -98,7 +99,7 @@ def _evaluate_local(test_data_path, hf_repo):
     dataloader = DataLoader(dataset, batch_size=8, shuffle=True)
 
 
-    model.eval()
+    global_model.eval()
     all_preds = []
     all_labels = []
 
@@ -107,7 +108,7 @@ def _evaluate_local(test_data_path, hf_repo):
             batch = {k: v.to(DEVICE) for k, v in batch.items()}
             labels = batch["labels"].type(torch.float32)
 
-            logits = model(batch["input_ids"], batch["attention_mask"])
+            logits = global_model(batch["input_ids"], batch["attention_mask"])
 
             preds = torch.sigmoid(logits).cpu().numpy() > 0.5
             labels = labels.cpu().numpy()

qwen_classifier/predict.py

@@ -1,7 +1,7 @@
 import torch
 import requests
 from .config import TAG_NAMES, SPACE_URL
-from .globals import model, tokenizer
+from .globals import global_model, global_tokenizer
 
 def predict_single(text, hf_repo, backend="local", hf_token=None):
     if backend == "local":
@@ -12,19 +12,19 @@ def predict_single(text, hf_repo, backend="local", hf_token=None):
         raise ValueError(f"Unknown backend: {backend}")
 
 def _predict_local(text, hf_repo):
-    global model, tokenizer
+    global global_model, global_tokenizer
     
     # Lazy-loading to avoid slow startup
-    if model is None:
+    if global_model is None:
         from .model import QwenClassifier
         from transformers import AutoTokenizer
         
-        model = QwenClassifier.from_pretrained(hf_repo).eval()
-        tokenizer = AutoTokenizer.from_pretrained(hf_repo)
+        global_model = QwenClassifier.from_pretrained(hf_repo).eval()
+        global_tokenizer = AutoTokenizer.from_pretrained(hf_repo)
     
-    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
+    inputs = global_tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
     with torch.no_grad():
-        logits = model(**inputs)
+        logits = global_model(**inputs)
     return _process_output(logits)
 
 def _predict_hf_api(text, hf_token=None):