changed app.py

app.py

@@ -2,15 +2,23 @@ from fastapi import FastAPI, Request, Form
 from fastapi.responses import HTMLResponse, RedirectResponse, JSONResponse
 from pydantic import BaseModel
 from typing import List
+import os                                    # ← add this
 from clearml import Model
 import torch
 from configs import add_args
 from models import build_or_load_gen_model
 import argparse
 from argparse import Namespace
-import os
 from peft import PeftModel, PeftConfig, get_peft_model, LoraConfig
 
+# ── Load ClearML secrets from HF Spaces environment ───────────────────────────
+CLEARML_WEB_SERVER   = os.environ["CLEARML_WEB_SERVER"]
+CLEARML_API_SERVER   = os.environ["CLEARML_API_SERVER"]
+CLEARML_FILES_SERVER = os.environ["CLEARML_FILES_SERVER"]
+CLEARML_ACCESS_KEY   = os.environ["CLEARML_API_ACCESS_KEY"]
+CLEARML_SECRET_KEY   = os.environ["CLEARML_API_SECRET_KEY"]
+# ───────────────────────────────────────────────────────────────────────────────
+
 MAX_SOURCE_LENGTH = 512
 
 def pad_assert(tokenizer, source_ids):
@@ -43,18 +51,24 @@ BASE_MODEL_NAME = "microsoft/codereviewer"
 args = Namespace(
     model_name_or_path=BASE_MODEL_NAME,
     load_model_path=None,
-    # Add other necessary default arguments if build_or_load_gen_model requires them
 )
 print(f"Loading base model architecture and tokenizer from: {BASE_MODEL_NAME}")
 config, base_model, tokenizer = build_or_load_gen_model(args)
 print("Base model architecture and tokenizer loaded.")
 
-# Download the fine-tuned weights from ClearML
-CLEARML_MODEL_ID = "34e25deb24c64b74b29c8519ed15fe3e"
-model_obj = Model(model_id=CLEARML_MODEL_ID)
+# Download the fine-tuned weights via ClearML using your injected creds
+model_obj = Model(
+    model_id="34e25deb24c64b74b29c8519ed15fe3e",
+    api_host=CLEARML_API_SERVER,
+    web_host=CLEARML_WEB_SERVER,
+    files_host=CLEARML_FILES_SERVER,
+    credentials={
+        "access_key": CLEARML_ACCESS_KEY,
+        "secret_key": CLEARML_SECRET_KEY,
+    },
+)
 finetuned_weights_path = model_obj.get_local_copy()
 adapter_dir = os.path.dirname(finetuned_weights_path)
-
 print(f"Fine-tuned adapter weights downloaded to directory: {adapter_dir}")
 
 # Create LoRA configuration matching the fine-tuned checkpoint