Update app.py

app.py

@@ -2,7 +2,7 @@ from fastapi import FastAPI, HTTPException
 from fastapi.middleware.cors import CORSMiddleware
 import torch
 import numpy as np
-from transformers import AutoTokenizer, AutoModel
+from transformers import AutoTokenizer, AutoModel, AutoConfig
 from typing import List, Union
 import json
 import logging
@@ -31,7 +31,18 @@ def load_model():
         logger.info(f"Loading Qwen3-Embedding-0.6B model on device: {DEVICE}")
         
         # Load tokenizer and model for Qwen3 embedding
-        tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
+        # First, try to load the config to understand the model structure
+        config = AutoConfig.from_pretrained(MODEL_NAME, trust_remote_code=True)
+        logger.info(f"Model config loaded: {config.model_type}")
+        
+        # Load tokenizer - try different approaches
+        try:
+            tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
+        except Exception as tokenizer_error:
+            logger.warning(f"Failed to load tokenizer with trust_remote_code=True: {tokenizer_error}")
+            tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=False)
+        
+        # Load model
         model = AutoModel.from_pretrained(
             MODEL_NAME, 
             trust_remote_code=True,
@@ -50,6 +61,7 @@ def load_model():
             test_output = model(**test_input)
             logger.info(f"Model test successful. Output shape: {test_output.last_hidden_state.shape}")
             logger.info(f"Model config hidden size: {model.config.hidden_size}")
+            logger.info(f"Tokenizer vocab size: {tokenizer.vocab_size}")
         
         logger.info("Qwen3-Embedding-0.6B model loaded successfully")
         return True