Spaces:
Running
Running
fixing token name
Browse files
app.py
CHANGED
@@ -12,7 +12,8 @@ from transformers import AutoTokenizer
|
|
12 |
from ddgs import DDGS
|
13 |
import spaces # Import spaces early to enable ZeroGPU support
|
14 |
|
15 |
-
|
|
|
16 |
|
17 |
# Optional: Disable GPU visibility if you wish to force CPU usage
|
18 |
# os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
@@ -139,7 +140,7 @@ def load_pipeline(model_name):
|
|
139 |
return PIPELINES[model_name]
|
140 |
repo = MODELS[model_name]["repo_id"]
|
141 |
tokenizer = AutoTokenizer.from_pretrained(repo,
|
142 |
-
token=access_token)
|
143 |
for dtype in (torch.bfloat16, torch.float16, torch.float32):
|
144 |
try:
|
145 |
pipe = pipeline(
|
@@ -150,7 +151,7 @@ def load_pipeline(model_name):
|
|
150 |
torch_dtype=dtype,
|
151 |
device_map="auto",
|
152 |
use_cache=False, # ← disable past-key-value caching
|
153 |
-
token=access_token)
|
154 |
PIPELINES[model_name] = pipe
|
155 |
return pipe
|
156 |
except Exception:
|
|
|
12 |
from ddgs import DDGS
|
13 |
import spaces # Import spaces early to enable ZeroGPU support
|
14 |
|
15 |
+
# Get Hugging Face token - works in both local and HF Spaces environments
|
16 |
+
access_token = os.environ.get('HF_TOKEN') or os.environ.get('HUGGINGFACE_HUB_TOKEN') or None
|
17 |
|
18 |
# Optional: Disable GPU visibility if you wish to force CPU usage
|
19 |
# os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
|
|
140 |
return PIPELINES[model_name]
|
141 |
repo = MODELS[model_name]["repo_id"]
|
142 |
tokenizer = AutoTokenizer.from_pretrained(repo,
|
143 |
+
token=access_token if access_token else None)
|
144 |
for dtype in (torch.bfloat16, torch.float16, torch.float32):
|
145 |
try:
|
146 |
pipe = pipeline(
|
|
|
151 |
torch_dtype=dtype,
|
152 |
device_map="auto",
|
153 |
use_cache=False, # ← disable past-key-value caching
|
154 |
+
token=access_token if access_token else None)
|
155 |
PIPELINES[model_name] = pipe
|
156 |
return pipe
|
157 |
except Exception:
|