roshnn24 commited on
Commit
0264bfa
·
verified ·
1 Parent(s): c188f54

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +10 -2
app.py CHANGED
@@ -31,6 +31,13 @@ app.config['UPLOAD_FOLDER'] = UPLOAD_FOLDER
31
  os.makedirs(UPLOAD_FOLDER, exist_ok=True)
32
  # Database configuration
33
  DATABASE_PATH = '/tmp/chat_database.db'
 
 
 
 
 
 
 
34
 
35
  # Initialize LangChain with Ollama LLM
36
  if hf_token:
@@ -40,12 +47,13 @@ else:
40
  model_name = "microsoft/phi-4"
41
 
42
  try:
43
- tokenizer = AutoTokenizer.from_pretrained(model_name)
44
  model = AutoModelForCausalLM.from_pretrained(
45
  model_name,
46
  torch_dtype=torch.float16,
47
  device_map="auto",
48
- load_in_8bit=True
 
49
  )
50
 
51
  # Create pipeline
 
31
  os.makedirs(UPLOAD_FOLDER, exist_ok=True)
32
  # Database configuration
33
  DATABASE_PATH = '/tmp/chat_database.db'
34
+ CACHE_DIR = "/tmp/huggingface_cache"
35
+ MODEL_CACHE_DIR = "/tmp/model_cache"
36
+ os.makedirs(CACHE_DIR, exist_ok=True)
37
+ os.makedirs(MODEL_CACHE_DIR, exist_ok=True)
38
+ os.environ['TRANSFORMERS_CACHE'] = CACHE_DIR
39
+ os.environ['HF_HOME'] = CACHE_DIR
40
+ os.environ['HF_DATASETS_CACHE'] = CACHE_DIR
41
 
42
  # Initialize LangChain with Ollama LLM
43
  if hf_token:
 
47
  model_name = "microsoft/phi-4"
48
 
49
  try:
50
+ tokenizer = AutoTokenizer.from_pretrained(model_name, cache_dir=CACHE_DIR)
51
  model = AutoModelForCausalLM.from_pretrained(
52
  model_name,
53
  torch_dtype=torch.float16,
54
  device_map="auto",
55
+ load_in_8bit=True,
56
+ cache_dir=MODEL_CACHE_DIR
57
  )
58
 
59
  # Create pipeline