Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
@@ -60,43 +60,34 @@ def test_llm_generation():
|
|
60 |
|
61 |
|
62 |
def initialize_model():
|
63 |
-
"""Separate model initialization with better error handling"""
|
64 |
global model, tokenizer, generation_config
|
65 |
-
|
66 |
try:
|
67 |
add_log("π Initializing model...")
|
68 |
-
|
69 |
tokenizer = AutoTokenizer.from_pretrained(
|
70 |
MODEL_ID,
|
71 |
trust_remote_code=True,
|
72 |
-
use_fast=False
|
73 |
)
|
74 |
-
|
75 |
-
# Ensure proper padding token
|
76 |
if tokenizer.pad_token is None:
|
77 |
tokenizer.pad_token = tokenizer.eos_token
|
78 |
add_log("β
Set pad_token to eos_token")
|
79 |
-
|
80 |
-
#
|
81 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
82 |
-
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
83 |
-
|
84 |
model = AutoModelForCausalLM.from_pretrained(
|
85 |
MODEL_ID,
|
86 |
-
torch_dtype=
|
87 |
-
device_map="auto" if torch.cuda.is_available() else None,
|
88 |
trust_remote_code=True,
|
|
|
89 |
low_cpu_mem_usage=True
|
90 |
)
|
91 |
-
|
92 |
-
if not torch.cuda.is_available():
|
93 |
-
model = model.to(device)
|
94 |
-
|
95 |
model.eval()
|
96 |
-
|
97 |
-
# Configure generation parameters
|
98 |
generation_config = GenerationConfig(
|
99 |
-
max_new_tokens=
|
100 |
temperature=0.7,
|
101 |
top_p=0.9,
|
102 |
do_sample=True,
|
@@ -105,9 +96,16 @@ def initialize_model():
|
|
105 |
repetition_penalty=1.1,
|
106 |
length_penalty=1.0
|
107 |
)
|
108 |
-
|
109 |
add_log(f"β
Model loaded successfully on device: {model.device}")
|
110 |
return True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
|
112 |
except Exception as e:
|
113 |
error_msg = f"β Model initialization failed: {str(e)}"
|
|
|
60 |
|
61 |
|
62 |
def initialize_model():
|
|
|
63 |
global model, tokenizer, generation_config
|
64 |
+
|
65 |
try:
|
66 |
add_log("π Initializing model...")
|
67 |
+
|
68 |
tokenizer = AutoTokenizer.from_pretrained(
|
69 |
MODEL_ID,
|
70 |
trust_remote_code=True,
|
71 |
+
use_fast=False
|
72 |
)
|
73 |
+
|
|
|
74 |
if tokenizer.pad_token is None:
|
75 |
tokenizer.pad_token = tokenizer.eos_token
|
76 |
add_log("β
Set pad_token to eos_token")
|
77 |
+
|
78 |
+
# Force GPU settings
|
|
|
|
|
|
|
79 |
model = AutoModelForCausalLM.from_pretrained(
|
80 |
MODEL_ID,
|
81 |
+
torch_dtype=torch.float16,
|
|
|
82 |
trust_remote_code=True,
|
83 |
+
device_map={"": 0}, # <- force GPU:0
|
84 |
low_cpu_mem_usage=True
|
85 |
)
|
86 |
+
|
|
|
|
|
|
|
87 |
model.eval()
|
88 |
+
|
|
|
89 |
generation_config = GenerationConfig(
|
90 |
+
max_new_tokens=4096,
|
91 |
temperature=0.7,
|
92 |
top_p=0.9,
|
93 |
do_sample=True,
|
|
|
96 |
repetition_penalty=1.1,
|
97 |
length_penalty=1.0
|
98 |
)
|
99 |
+
|
100 |
add_log(f"β
Model loaded successfully on device: {model.device}")
|
101 |
return True
|
102 |
+
|
103 |
+
except Exception as e:
|
104 |
+
error_msg = f"β Model initialization failed: {str(e)}"
|
105 |
+
add_log(error_msg)
|
106 |
+
add_log(f"Traceback: {traceback.format_exc()}")
|
107 |
+
return False
|
108 |
+
|
109 |
|
110 |
except Exception as e:
|
111 |
error_msg = f"β Model initialization failed: {str(e)}"
|