model problem solved

App/app.py

@@ -5,7 +5,6 @@ from functools import wraps
 from supabase import create_client
 from flask_jwt_extended import JWTManager, create_access_token,unset_jwt_cookies, jwt_required, get_jwt_identity,decode_token
 from App.models import  User,LostItem,FoundItem,Match
-from dotenv import load_dotenv
 from flask_cors import CORS
 from src.training import encode_img_and_text
 from qdrant_client import QdrantClient
@@ -20,7 +19,6 @@ from datetime import timedelta
 warnings.filterwarnings("ignore", message=".*QuickGELU mismatch.*")
 
 
-load_dotenv()
 
 app = Flask(__name__)
 app.config["SQLALCHEMY_DATABASE_URI"] = os.getenv("DATABASE_URL")

allfiles.txt

@@ -0,0 +1,32 @@
+2fc8608ffc5292bfd2ea1bbf97b7e50c0ca797ac
+6cca4d3a19bbdbbd248b38fefcffb6e799115e4a
+8a58eaa680084c7828b7190d72a28a27657e40f5
+a718cdd3df5c477a331c5658ee7243b473e28711
+d24b09fbb595bc12d5edb593a773cfd84888c831
+26ecb4581bef71354473073c50f0114adee83454 
+32212c33b8a43bacaafa60517b709500202868e0 
+40dd086edb25acc272d44312232b432a65b3cf92 
+4f7c84bb828dc03ace4e24a1d9b4a075b814f966 
+b542cfcc9811103dcb34600f304fef079586c745 
+315969e673bb1c2280258455ee10fe654cef7e58 .env
+a6344aac8c09253b3b630fb776ae94478aa0275b .gitattributes
+354e4823dd1ab296ac6f2a5109a80019bd3cffd6 .gitignore
+de6c6c2d8cb15165b4e8f7ab22f58dfc40fd3f1f .gitignore
+c4e737bc10b777de9ff0b3d6412c67f833e3e285 App
+e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 App/__init__.py
+025c75b2d2b2df29802ed1f3b707e999fe15ac41 App/app.py
+bcc7d77ffc45f7c1cce1ff2520e29b3f1196d2e3 App/models.py
+7108ef9623bce7e41f86ed1975f02e545bcf84ff App/scheduler.py
+fecba207f6b5c465c85039cdc28640b82772b0d8 Dockerfile
+f5824832d73309e87386f99ebca332120d0b6e4f model
+b9aa3f250bd0cdb769d753e0847ac454b03b258b model/clip
+116a8bb85040164dcf8dfde96112b331857465c1 model/clip/best.pt
+04135bae2a3d5f95ac25a5f7117a822516b4a9ef model/clip/epoch5_val0.7777.pt
+7b064d8f8f7e759fc42f24447a2ba2e625f77a73 README.md
+2f808aa33de8a014b003f0c5430f217fa79ca876 requirements.txt
+5d2e66968630adbd91f3c44caeb34446b44f9360 requirements_scheduler.txt
+6ad0ff8380d2b4136fe12becdf1928fd301778b7 setup.py
+edfc8a8148ea12a4982caf45fc2a3c3c7b84dfe5 src
+6bd3ca06c394c2095f295e6889d7f706005d40b5 src/evaluation.py
+8cd3fa1c487903893ff4c3f5c40495c0dc3f9c1d src/preprocessing.py
+25d7f73ad394a323774f913e50315fb5d3832a2b src/training.py

src/evaluation.py

@@ -6,9 +6,19 @@ import open_clip
 from datasets import load_dataset
 from tqdm import tqdm
 import numpy as np
+from huggingface_hub import hf_hub_download
 
 import warnings
 warnings.filterwarnings("ignore", message=".*QuickGELU mismatch.*")
+HF_TOKEN=os.getenv("HF_TOKEN")
+
+MODEL_ID = "PrashantGoyal/findr-clip-ft"
+
+model_path = hf_hub_download(
+    repo_id=MODEL_ID,
+    filename="best.pt",
+    token=os.getenv("HF_TOKEN")
+)
 
 def collate(batch):
     img,text=zip(*batch)
@@ -43,7 +53,7 @@ def main(path='./model/clip/best.pt',arch='ViT-B-32', pretrained='openai'):
     torch.cuda.empty_cache()
     model, _, preprocess =open_clip.create_model_and_transforms(arch,pretrained=pretrained,device=device,quick_gelu=True )
     tokenizer=open_clip.get_tokenizer(arch)
-    state=torch.load(path,map_location='cuda')['model']
+    state=torch.load(model_path,map_location='cuda')['model']
     model.load_state_dict(state, strict=False)
     model.eval()
     print('model loaded')

src/training.py

@@ -10,13 +10,22 @@ from src.preprocessing import Preprocessing
 from torch.utils.data import DataLoader,Dataset
 import warnings
 import base64
+from huggingface_hub import hf_hub_download
 from io import BytesIO
 warnings.filterwarnings("ignore", message=".*QuickGELU mismatch.*")
 
 device='cuda' if torch.cuda.is_available() else 'cpu'
 torch.cuda.empty_cache()
 model, _, preprocess =open_clip.create_model_and_transforms('ViT-B-32',pretrained='openai',device=device )
-SAVE_DIR='model/clip/best.pt'
+HF_TOKEN=os.getenv("HF_TOKEN")
+
+MODEL_ID = "PrashantGoyal/findr-clip-ft"
+
+model_path = hf_hub_download(
+    repo_id=MODEL_ID,
+    filename="best.pt",
+    token=os.getenv("HF_TOKEN")
+)
 tokenizer=open_clip.get_tokenizer('ViT-B-32')
 
 def seed_everything(seed=42):
@@ -150,7 +159,7 @@ def feedback(model,processor,device,data,epochs=5,batch_size=4,lr=1e-6):
     dataLoader=DataLoader(dataset,batch_size=batch_size,shuffle=True)
     optimizer = torch.optim.AdamW(model.parameters(), lr=lr)
     loss_fn = nn.CosineEmbeddingLoss()
-    model.load_state_dict(torch.load(SAVE_DIR, map_location=device))
+    model.load_state_dict(torch.load(model_path, map_location=device))
     model.train()
     for epoch in range(epochs):
         total_loss = 0
@@ -178,7 +187,7 @@ def feedback(model,processor,device,data,epochs=5,batch_size=4,lr=1e-6):
 def encode_img_and_text(imgs,text):
     image_feat=[]
     model, _, preprocess =open_clip.create_model_and_transforms('ViT-B-32',pretrained='openai',device=device,quick_gelu=True )
-    checkpoint = torch.load(SAVE_DIR, map_location=device)
+    checkpoint = torch.load(model_path, map_location=device)
     model.to(device)
     for img in imgs:
         if hasattr(img, 'read'):