Update app.py
Browse files
app.py
CHANGED
@@ -22,7 +22,7 @@ USE_SYSTEM_PROMPT = os.environ.get('USE_SYSTEM_PROMPT', '').lower() == "true" or
|
|
22 |
ENABLE_GPU = os.environ.get('ENABLE_GPU', '').lower() == "true" or False
|
23 |
GPU_LAYERS = int(os.environ.get('GPU_LAYERS', '0'))
|
24 |
n_gqa_str = os.environ.get('N_GQA')
|
25 |
-
N_GQA = int(n_gqa_str) if n_gqa_str is not None else None #must be set to 8 for 70b models
|
26 |
CHAT_FORMAT = os.environ.get('CHAT_FORMAT') or 'llama-2'
|
27 |
REPO_NAME = os.environ.get('REPO_NAME') or 'IlyaGusev/saiga2_7b_gguf'
|
28 |
MODEL_NAME = os.environ.get('MODEL_NAME') or 'model-q4_K.gguf'
|
|
|
22 |
ENABLE_GPU = os.environ.get('ENABLE_GPU', '').lower() == "true" or False
|
23 |
GPU_LAYERS = int(os.environ.get('GPU_LAYERS', '0'))
|
24 |
n_gqa_str = os.environ.get('N_GQA')
|
25 |
+
N_GQA = 8#int(n_gqa_str) if n_gqa_str is not None else None #must be set to 8 for 70b models
|
26 |
CHAT_FORMAT = os.environ.get('CHAT_FORMAT') or 'llama-2'
|
27 |
REPO_NAME = os.environ.get('REPO_NAME') or 'IlyaGusev/saiga2_7b_gguf'
|
28 |
MODEL_NAME = os.environ.get('MODEL_NAME') or 'model-q4_K.gguf'
|