MGZON commited on
Commit
ce84fea
·
verified ·
1 Parent(s): c6b5a6a

Fix cache path and permissions for T5 model loading

Browse files
Files changed (1) hide show
  1. app.py +11 -14
app.py CHANGED
@@ -10,9 +10,8 @@ logging.basicConfig(level=logging.INFO)
10
  logger = logging.getLogger(__name__)
11
 
12
  # Set up cache directory
13
- CACHE_DIR = os.environ.get("HF_HOME", "/app/.cache/huggingface")
14
-
15
- # Ensure libraries use the cache directory
16
  os.environ["HF_HOME"] = CACHE_DIR
17
 
18
  # Create the FastAPI app
@@ -27,25 +26,23 @@ async def health_check():
27
  return {"status": "healthy"}
28
 
29
  # Load T5 model from local cache
30
- T5_REPO = "MGZON/mgzon-flan-t5-base"
31
  try:
32
- logger.info(f"Loading tokenizer for {T5_REPO} from local cache")
33
  t5_tokenizer = AutoTokenizer.from_pretrained(
34
- T5_REPO,
35
- cache_dir=CACHE_DIR,
36
  local_files_only=True
37
  )
38
- logger.info(f"Successfully loaded tokenizer for {T5_REPO}")
39
- logger.info(f"Loading model for {T5_REPO} from local cache")
40
  t5_model = AutoModelForSeq2SeqLM.from_pretrained(
41
- T5_REPO,
42
- cache_dir=CACHE_DIR,
43
  local_files_only=True
44
  )
45
- logger.info(f"Successfully loaded model for {T5_REPO}")
46
  except Exception as e:
47
- logger.error(f"Failed to load T5 model from {T5_REPO}: {str(e)}")
48
- raise RuntimeError(f"Failed to load T5 model from {T5_REPO}: {str(e)}")
49
 
50
  # Load Mistral GGUF model
51
  gguf_path = os.path.abspath("models/mistral-7b-instruct-v0.1.Q4_K_M.gguf")
 
10
  logger = logging.getLogger(__name__)
11
 
12
  # Set up cache directory
13
+ CACHE_DIR = "/app/.cache/huggingface/hub"
14
+ os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
 
15
  os.environ["HF_HOME"] = CACHE_DIR
16
 
17
  # Create the FastAPI app
 
26
  return {"status": "healthy"}
27
 
28
  # Load T5 model from local cache
29
+ T5_MODEL_PATH = os.path.join(CACHE_DIR, "models--MGZON--mgzon-flan-t5-base/snapshots")
30
  try:
31
+ logger.info(f"Loading tokenizer for MGZON/mgzon-flan-t5-base from {T5_MODEL_PATH}")
32
  t5_tokenizer = AutoTokenizer.from_pretrained(
33
+ T5_MODEL_PATH,
 
34
  local_files_only=True
35
  )
36
+ logger.info(f"Successfully loaded tokenizer for MGZON/mgzon-flan-t5-base")
37
+ logger.info(f"Loading model for MGZON/mgzon-flan-t5-base from {T5_MODEL_PATH}")
38
  t5_model = AutoModelForSeq2SeqLM.from_pretrained(
39
+ T5_MODEL_PATH,
 
40
  local_files_only=True
41
  )
42
+ logger.info(f"Successfully loaded model for MGZON/mgzon-flan-t5-base")
43
  except Exception as e:
44
+ logger.error(f"Failed to load T5 model from {T5_MODEL_PATH}: {str(e)}")
45
+ raise RuntimeError(f"Failed to load T5 model from {T5_MODEL_PATH}: {str(e)}")
46
 
47
  # Load Mistral GGUF model
48
  gguf_path = os.path.abspath("models/mistral-7b-instruct-v0.1.Q4_K_M.gguf")